CN113314092A - Method and device for model training and voice interaction - Google Patents
Method and device for model training and voice interaction Download PDFInfo
- Publication number
- CN113314092A CN113314092A CN202110509888.1A CN202110509888A CN113314092A CN 113314092 A CN113314092 A CN 113314092A CN 202110509888 A CN202110509888 A CN 202110509888A CN 113314092 A CN113314092 A CN 113314092A
- Authority
- CN
- China
- Prior art keywords
- voice
- voice data
- information
- data
- sound source
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000012549 training Methods 0.000 title claims abstract description 51
- 230000003993 interaction Effects 0.000 title claims abstract description 30
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 106
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 106
- 238000005457 optimization Methods 0.000 claims abstract description 9
- 238000012216 screening Methods 0.000 claims description 21
- 238000004590 computer program Methods 0.000 claims description 17
- 238000003860 storage Methods 0.000 claims description 16
- 230000009747 swallowing Effects 0.000 claims description 9
- 230000004044 response Effects 0.000 claims description 5
- 238000010586 diagram Methods 0.000 description 16
- 230000008569 process Effects 0.000 description 11
- 230000006870 function Effects 0.000 description 10
- 230000006872 improvement Effects 0.000 description 9
- 238000012545 processing Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 230000003287 optical effect Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 206010019133 Hangover Diseases 0.000 description 1
- 241000167880 Hirundinidae Species 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 229920001296 polysiloxane Polymers 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L13/00—Speech synthesis; Text to speech systems
- G10L13/02—Methods for producing synthetic speech; Speech synthesisers
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L13/00—Speech synthesis; Text to speech systems
- G10L13/02—Methods for producing synthetic speech; Speech synthesisers
- G10L13/033—Voice editing, e.g. manipulating the voice of the synthesiser
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L13/00—Speech synthesis; Text to speech systems
- G10L13/08—Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/26—Speech to text systems
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/78—Detection of presence or absence of voice signals
- G10L25/87—Detection of discrete points within a voice signal
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
- G10L2015/025—Phonemes, fenemes or fenones being the recognition units
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Electrically Operated Instructional Devices (AREA)
Abstract
The specification discloses a method and a device for model training and voice interaction, wherein a service platform can acquire voice data and determine to-be-compensated voice information corresponding to the voice data according to the voice data, wherein the to-be-compensated voice information is used for representing original voice characteristics corresponding to each voice unit contained in the voice data. And finally, training the voice synthesis model by taking the minimized deviation between the voice synthesis result and the voice data as an optimization target, thereby carrying out voice synthesis through the voice synthesis model and reducing the deviation between the automatically synthesized voice and the voice spoken by people at ordinary times.
Description
Technical Field
The present disclosure relates to the field of machine learning, and in particular, to a method and an apparatus for model training and speech interaction.
Background
With the continuous development of information technology, services such as intelligent voice customer service and voice navigation, which rely on automatic voice synthesis, have gradually been integrated into the lives of people, and convenience is brought to the lives of people.
In these services, how to automatically synthesize the speech is the key point for better implementing these services, and in the prior art, the automatically synthesized speech usually deviates greatly from the ordinary speaking of people, for example, some automatically synthesized speech is more mechanized, the time interval of each spoken word is usually more consistent, and if the speech is applied to the services such as the intelligent speech service, the speech navigation, etc., the user may have difficulty in adapting to the speech.
Therefore, how to reduce the deviation between the automatically synthesized speech and the speech spoken by people at ordinary times is an urgent problem to be solved.
Disclosure of Invention
The present specification provides a method and apparatus for model training and speech interaction to partially solve the above problems in the prior art.
The technical scheme adopted by the specification is as follows:
the present specification provides a method of model training, comprising:
acquiring voice data;
according to the voice data, determining voice information to be compensated corresponding to the voice data, wherein the voice information to be compensated is used for representing original voice characteristics corresponding to all voice units contained in the voice data;
determining pronunciation habit characteristics corresponding to the voice data, and compensating the voice information to be compensated according to the pronunciation habit characteristics to obtain compensated voice information;
inputting the compensated voice information into a voice synthesis model to be trained to obtain a voice synthesis result;
and training the voice synthesis model by taking the minimized deviation between the voice synthesis result and the voice data as an optimization target.
Optionally, determining, according to the voice data, to-be-compensated voice information corresponding to the voice data, specifically including:
segmenting the voice data to obtain clauses, wherein each clause comprises at least one voice unit;
determining the voice information to be compensated corresponding to each clause;
and determining the voice information to be compensated corresponding to the voice data according to the voice information to be compensated corresponding to each clause.
Optionally, determining, according to the voice data, to-be-compensated voice information corresponding to the voice data, specifically including:
determining at least one sound source in the voice data, and determining audio data corresponding to each sound source in the at least one sound source according to the voice data;
selecting a target sound source from the at least one sound source according to the audio data corresponding to each sound source in the at least one sound source;
and determining the voice information to be compensated corresponding to the voice data according to the audio data corresponding to the target sound source.
Optionally, selecting a target sound source from the at least one sound source according to the audio data corresponding to each sound source in the at least one sound source, specifically including:
determining the integral voiceprint characteristics corresponding to the at least one sound source and determining the sound source voiceprint characteristics corresponding to each sound source contained in the at least one sound source according to the audio data corresponding to each sound source in the at least one sound source;
and selecting a target sound source from the at least one sound source according to the overall voiceprint characteristics and the sound source voiceprint characteristics corresponding to each sound source contained in the at least one sound source.
Optionally, determining, according to the voice data, to-be-compensated voice information corresponding to the voice data, specifically including:
screening out voice data with voice quality meeting preset conditions from the voice data to serve as target voice data;
and determining the voice information to be compensated corresponding to the voice data according to the target voice data.
Optionally, the screening, from the voice data, the voice data whose voice quality meets a preset condition as the target voice data specifically includes:
screening out voice data with a signal-to-noise ratio not lower than a set signal-to-noise ratio from the voice data as target voice data; or
Screening out voice data with the voice amplitude within a set amplitude range from the voice data to serve as target voice data; or
Screening out voice data with the fundamental frequency within a set fundamental frequency range from the voice data to serve as target voice data; or
Screening out voice data with the voice duration not lower than the set duration from the voice data as target voice data; or
And screening out the voice data with the speed of speech within a set speed of speech range from the voice data as target voice data.
Optionally, the speech unit includes: a phoneme; the voice information to be compensated comprises at least one of phoneme information and original pause information, and the original pause information is used for representing an original position of a voice pause in the voice data; the pronunciation habit features include: at least one of a pause feature, a drag feature, and a swallow feature.
Optionally, determining the pronunciation habit characteristics corresponding to the voice data specifically includes:
and for each phoneme contained in the voice data, if the duration corresponding to the phoneme exceeds a first set duration, determining that the pronunciation habit feature corresponding to the phoneme is a dragging feature, and if the duration corresponding to the phoneme is less than a second set duration, determining that the pronunciation habit feature corresponding to the phoneme is a swallowing feature, wherein the second set duration is shorter than the first set duration.
The present specification provides a method of voice interaction, comprising:
responding to a voice interaction request of a user, and determining a voice message sent by the user;
determining text information replied aiming at the voice message according to the voice message and a preset reply strategy;
converting the text information into voice synthesis information, inputting the voice synthesis information into a pre-trained voice synthesis model to obtain a synthesized voice message, wherein the voice synthesis information is used for representing voice characteristics corresponding to each voice unit contained in the text information, and the voice synthesis model is obtained by training through a model training method;
and feeding back the synthesized voice message to the user.
The present specification provides an apparatus for model training, comprising:
the acquisition module is used for acquiring voice data;
the information determining module is used for determining to-be-compensated voice information corresponding to the voice data according to the voice data, wherein the to-be-compensated voice information is used for representing original voice characteristics corresponding to each voice unit contained in the voice data;
the characteristic determining module is used for determining pronunciation habit characteristics corresponding to the voice data and compensating the voice information to be compensated according to the pronunciation habit characteristics to obtain compensated voice information;
the input module is used for inputting the compensated voice information into a voice synthesis model to be trained to obtain a voice synthesis result;
and the training module is used for training the voice synthesis model by taking the minimized deviation between the voice synthesis result and the voice data as an optimization target.
This specification provides a voice interaction device, comprising:
the response module is used for responding to a voice interaction request of a user and determining a voice message sent by the user;
the text determining module is used for determining text information replied aiming at the voice message according to the voice message and a preset reply strategy;
the input module is used for converting the text information into voice synthesis information and inputting the voice synthesis information into a pre-trained voice synthesis model to obtain a synthesized voice message;
and the feedback module is used for feeding back the synthesized voice message to the user, the voice synthesis information is used for representing the voice characteristics corresponding to each voice unit contained in the text information, and the voice synthesis model is obtained by training through a model training method.
The present specification provides a computer-readable storage medium, which stores a computer program that, when executed by a processor, implements the above-described method of model training or speech interaction.
The present specification provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the above method of model training or speech interaction when executing the program.
The technical scheme adopted by the specification can achieve the following beneficial effects:
in the method and apparatus for model training and voice interaction provided in this specification, a service platform may obtain voice data, and determine, according to the voice data, to-be-compensated voice information corresponding to the voice data, where the to-be-compensated voice information is used to represent original voice features corresponding to each voice unit included in the voice data. And then, determining pronunciation habit characteristics corresponding to the voice data, compensating the voice information to be compensated according to the pronunciation habit characteristics to obtain compensated voice information, inputting the compensated voice information into a voice synthesis model to be trained to obtain a voice synthesis result, and training the voice synthesis model by taking the deviation between the voice synthesis result and the voice data as an optimization target, so that the voice synthesis model can be applied to various services related to voice synthesis in the subsequent process.
It can be seen from the above method that, during model training, the service platform can compensate the voice information to be compensated showing the original voice characteristics, so that the compensated voice information can show the individual pronunciation habits of the speakers in the voice data, and after model training, the voice synthesis model can be synthesized no matter what pronunciation habits the service platform needs, so that the voice synthesis model can synthesize the voice conforming to the ordinary speaking habits of people.
Drawings
The accompanying drawings, which are included to provide a further understanding of the specification and are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and together with the description serve to explain the specification and not to limit the specification in a non-limiting sense. In the drawings:
FIG. 1 is a schematic flow chart of a method of model training in the present specification;
fig. 2 is a schematic diagram illustrating compensation of speech information to be compensated provided in this specification;
FIG. 3 is a flow chart illustrating a method of voice interaction in the present specification;
FIG. 4 is a schematic diagram of an apparatus for model training in the present specification;
FIG. 5 is a schematic diagram of a voice interaction apparatus;
fig. 6 is a schematic diagram of an electronic device corresponding to fig. 1 or fig. 3 provided in the present specification.
Detailed Description
In order to make the objects, technical solutions and advantages of the present disclosure more clear, the technical solutions of the present disclosure will be clearly and completely described below with reference to the specific embodiments of the present disclosure and the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present disclosure, and not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present specification without any creative effort belong to the protection scope of the present specification.
The technical solutions provided by the embodiments of the present description are described in detail below with reference to the accompanying drawings.
Fig. 1 is a schematic flow chart of a model training method in this specification, which specifically includes the following steps:
s101: voice data is acquired.
S102: and determining to-be-compensated voice information corresponding to the voice data according to the voice data, wherein the to-be-compensated voice information is used for representing original voice characteristics corresponding to each voice unit contained in the voice data.
In practical application, if a service platform needs to build a service related to speech synthesis, a speech synthesis model needs to be built, trained and the like, and the trained speech synthesis model can generate corresponding speech according to a text given by the service platform, so that the service platform can apply the speech synthesis model to the service related to speech synthesis. In the training phase of the speech synthesis model, a large amount of data for training the model is required.
Based on this, the service platform can obtain the voice data and determine the voice information to be compensated corresponding to the voice data according to the voice data. The voice data can be obtained in various ways, such as recording in the process of manual customer service conversation, recording of ordinary speaking of people, and the like. The voice information to be compensated is used to represent the original voice characteristics corresponding to each voice unit contained in the voice data, that is, no matter which person in the voice data says, the original voice characteristics are consistent for the same words (with the same characters), that is, the voice information to be compensated represents the common speaking habit.
The above-mentioned speech units may refer to phonemes, i.e., minimum speech units. The above-mentioned speech information to be compensated may include phoneme information, original pause information, and tone information, where the tone information is used to indicate a level and a narrow pitch in the speech data, and the original pause information is used to indicate an original position of a speech pause in the speech data. That is, the original pause information indicates a pause position to be occurred when the text corresponding to the voice data is spoken. The phoneme information may indicate phonemes corresponding to texts of the voice data. Of course, the phonetic units referred to herein may also be words.
Because the speech information to be compensated represents a common speaking habit, when determining the speech information to be compensated, the text corresponding to the speech data may be determined, and then the speech information to be compensated, such as the above-mentioned phoneme information and original pause information, may be determined by the text corresponding to the speech data, that is, the text is converted into phonemes to determine the phoneme information corresponding to the speech data. The original pause information can be determined according to the pause position where the text corresponding to the voice data should appear. The pause position where the text corresponding to the voice data should appear can be determined according to the semantics of the text. For example, assuming that the text corresponding to the voice data is "bloomed in park of city a", since city a is a phrase, the pause position to be occurred can be determined after city a, so that it is determined that the original pause information corresponding to the text indicates that there is a pause position after city a.
The voice data mentioned above may refer to a plurality of voice data with a short duration, that is, each voice data includes a speech of a sentence, and of course, the voice data may also be voice data with a long duration and a relatively complete speech. In order to train the speech synthesis model, if the speech data is speech data with a long duration, the speech data may be segmented to obtain each clause, where each clause includes at least one speech unit, and the service platform may determine, for each clause, speech information to be compensated corresponding to the clause, and determine, according to the speech information to be compensated corresponding to each clause, speech information to be compensated corresponding to the speech data. That is to say, the Voice data is divided into a plurality of short sentences, and the to-be-compensated Voice information corresponding to each short sentence is determined as the to-be-compensated Voice information corresponding to the Voice data, and specifically, the plurality of short sentences in the Voice data can be determined by Voice endpoint Detection (VAD).
It should be noted that the voice data may include different sound sources, that is, sounds of different people, and in practical applications, only voices of the same sound source may need to be synthesized, so that the service platform may determine at least one sound source in the voice data and determine, according to the voice data, audio data corresponding to each sound source in the at least one sound source, thereby selecting a target sound source from the at least one sound source according to the audio data corresponding to each sound source in the at least one sound source, and determining, according to the audio data corresponding to the target sound source, to-be-compensated voice information corresponding to the voice data.
That is to say, the service platform needs to determine audio data corresponding to one sound source, that is, audio data corresponding to a target sound source, so as to train the speech synthesis model through the audio data of the target sound source, and thus speech synthesized by the trained speech synthesis model is sound of the target sound source.
It should be noted that, there may be multiple ways of selecting and extracting the target sound source from the at least one sound source, for example, the service platform may determine, according to the audio data corresponding to each sound source in the at least one sound source, an overall voiceprint feature corresponding to the at least one sound source, determine, from the at least one sound source, a sound source voiceprint feature corresponding to each sound source included in the at least one sound source, and select and extract the target sound source from the at least one sound source according to the overall voiceprint feature and the sound source voiceprint feature corresponding to each sound source.
That is, the above-described overall voiceprint feature is determined from the complete speech data, and the overall voiceprint feature reflects the voiceprint of the overall speech data. The voice data may be voice data corresponding to a plurality of phrases, each phrase corresponds to a sound source, one sound source may correspond to a plurality of phrases, and the service platform may extract voiceprint features for each phrase to obtain the voiceprint features of the sound source corresponding to the phrase.
The service platform can select a short sentence with the similarity between the sound source voiceprint characteristics and the overall voiceprint characteristics higher than the set similarity, and train the voice synthesis model through the voice data corresponding to the short sentence, namely, the sound source selected by the service platform is the sound source which accounts for the most in the overall voice data and serves as the target sound source. Of course, the service platform may also select other sound sources, and the service platform may select a required sound source according to actual requirements, for example, select a sound source that occupies the smallest ratio in the entire voice data.
It should be further noted that, in order to ensure the quality of the speech synthesized by the speech synthesis model, the speech quality of the speech data used for training the speech synthesis model needs to be ensured, so that the service platform may screen out the speech data whose speech quality meets the preset condition from the speech data as target speech data, and determine the to-be-compensated speech information corresponding to the speech data according to the target speech data. The voice data acquired by the service platform may include a section of voice, and certainly, the voice data may also be multiple, each voice data corresponds to a voice of a short sentence, and when the service platform is used for screening, a part of voice may be screened from a section of voice, and a plurality of voice data may also be screened from a plurality of voice data, and the selected voice data is used as target voice data.
The preset conditions mentioned above may be preset, and there may be multiple preset conditions, that is, there may be multiple ways for the service platform to screen the target voice data from the voice data. For example, the service platform may screen out, from the voice data, voice data having a signal-to-noise ratio not lower than a set signal-to-noise ratio as target voice data. The signal-to-noise ratio refers to a ratio between a voice signal and noise, that is, the service platform needs to select voice data with a higher ratio of the voice signal to noise such as environmental noise and call noise.
For another example, the service platform may filter out, from the voice data, voice data with a sound amplitude within a set amplitude range, as target voice data, where the sound amplitude represents the volume of the voice in the voice data, that is, it is not desirable to train the voice synthesis model with voice data with too low volume, nor to train the voice synthesis model with voice data with too high volume, for example, it is undesirable to train the voice synthesis model with voice data with a degree that the voice may reach a microphone.
For another example, the service platform may filter out, from the voice data, voice data whose fundamental frequency is within a set fundamental frequency range as target voice data, where the fundamental frequency mentioned here represents the pitch of the sound in the voice data, that is, it is desirable that the pitch of the target voice data selected can be appropriate, and the sound in the target voice data is neither too sharp nor too deep.
For another example, the voice data may be voice data of multiple phrases, and if the duration of the voice data corresponding to one phrase is too short, the voice data corresponding to the phrase may not contain much voice, so that the service platform may screen out the voice data having a voice duration not less than the set duration from the voice data as the target voice data.
For another example, the service platform may screen out, from the voice data, the voice data whose voice speed is within the set voice speed range as target voice data, that is, the voice synthesis model is not trained by the voice data whose voice speed is too fast or too slow, and the above-mentioned set signal-to-noise ratio, set amplitude range, set fundamental frequency range, set duration and the set voice speed range mentioned here may all be preset.
Through the series of screening operations, the screened target voice data can be ensured to have higher voice quality, so that the quality of the voice data acquired by the service platform does not need to be limited, namely, the voice data recorded in a specific recording environment, such as the voice data recorded in a recording studio, does not need to be acquired. The voice data obtained by the service platform through the modes of conventional conversation, recording by common recording equipment and the like can be applied to the method, so that the cost can be reduced to a certain extent.
S103: and determining pronunciation habit characteristics corresponding to the voice data, and compensating the voice information to be compensated according to the pronunciation habit characteristics to obtain compensated voice information.
The service platform determines voice information to be compensated corresponding to the voice data, can determine pronunciation habit characteristics corresponding to the voice data, and compensates the voice information to be compensated according to the pronunciation habit characteristics to obtain compensated voice information.
The pronunciation habit feature mentioned here can indicate the pronunciation habit of the speaker person in the voice data, not the common pronunciation habit, such as the pause of the speaker person's own habit when speaking, the dragging habit of the speaker's own, and so on. Accordingly, the pronunciation habit features may include a pause feature, a tug feature, a swallow feature, and the like. The pause feature is used to characterize the location of pauses in speech in the speech data due to the individual habits of the speaker. Similar are the hangover feature, which characterizes sound locations in the speech data that are elongated due to the personal habits of the speaker, and the gulp feature, which characterizes sound locations in the speech data that are more transient due to the personal habits of the speaker.
Because the swallowing characteristic and the dragging characteristic are related to the length of the sound, when the service platform determines the swallowing characteristic and the dragging characteristic, the service platform may determine, for each phoneme in the speech data, a duration corresponding to the phoneme, determine, if the duration corresponding to the phoneme exceeds a first set duration, that the pronunciation habit characteristic corresponding to the phoneme is the dragging characteristic, and if the pronunciation habit characteristic corresponding to the phoneme is the swallowing characteristic, where the second set duration is shorter than the first set duration, that is, the second set duration is shorter and the first set duration is longer. When the pause feature is determined, the service platform can determine the position where the voice does not appear in the voice data, and mark the position as the pause position.
When the duration corresponding to the phoneme is determined, the duration corresponding to the phoneme can be determined through the voice data and the text corresponding to the voice data, that is, when the voice data is converted into the text, each character can correspond to a duration in the voice data, and the phoneme can also correspond to the character, so that the duration corresponding to the phoneme can be determined through the duration corresponding to the character.
After determining the pronunciation habit characteristics corresponding to the voice data, the service platform can compensate the voice information to be compensated according to the pronunciation habit characteristics to obtain compensated voice information. For example, for the phoneme information in the speech information to be compensated, if the pronunciation habit feature corresponding to a certain phoneme in the phoneme information is the lingering feature, the symbol of the phoneme in the phoneme information may be changed to the lingering phoneme symbol corresponding to the phoneme, and if the pronunciation habit feature corresponding to a certain phoneme in the phoneme information is the swallowing feature, the symbol of the phoneme in the phoneme information may be changed to the swallowing phoneme symbol corresponding to the phoneme, so as to obtain the compensated speech information, as shown in fig. 2.
Fig. 2 is a schematic diagram of compensating for speech information to be compensated provided in this specification.
As can be seen from fig. 2, the speech information to be compensated is "zh ong u y ong g d u" obtained by directly converting the text into phonemes, and in the speech data, the voice of the speaker in the speech data is elongated when speaking some phonemes, so that the pronunciation habit characteristics corresponding to these phonemes can be determined as the lingering characteristics, then when determining the speech information to be compensated, the service platform can replace the symbols of these phonemes in the speech information, and the lingering phoneme symbols corresponding to the corresponding phonemes, so as to obtain the compensated speech information, for example, d in the speech information to be compensated in fig. 2 is replaced to obtain the compensated speech informationdU is replaced byuSimilarly, the swallowing may be represented by another symbol, so that the speech synthesis model can determine which phonemes are dragons and which are swallows when synthesizing the speech.
For the pause feature, the original pause information may be modified according to the pause feature to obtain the compensated pause information, and the above-mentioned "spending in park in city a" is referred to continuously, where the original pause information indicates the pause position after city a, and the speaker in the speech data actually pauses after city a, but does not pause after city, the compensated pause information may indicate the pause position after city a.
S104: and inputting the compensated voice information into a voice synthesis model to be trained to obtain a voice synthesis result.
S105: and training the voice synthesis model by taking the minimized deviation between the voice synthesis result and the voice data as an optimization target.
And the service platform inputs the compensated voice information into a voice synthesis model to be trained to obtain a voice synthesis result, and trains the voice synthesis model by taking the minimized deviation between the voice synthesis result and the voice data as an optimization target.
When model training is carried out, the service platform can extract actual audio features corresponding to voice data, and after compensated voice information is input into a voice synthesis model to be trained, audio features synthesized by the voice synthesis model can be obtained, and the voice synthesis model is trained by taking the aim of minimizing the deviation between the audio features synthesized by the voice synthesis model and the actual audio features. The actual audio features mentioned here are obtained from the frequency domain features corresponding to the speech data.
That is, the purpose of training the speech synthesis model is to learn the mapping relationship between the compensated speech information and the corresponding audio features, that is, the trained speech synthesis model can synthesize the corresponding audio features by the information similar to the compensated speech information, so that the service platform obtains the synthesized speech.
It should be noted that, as mentioned above, the voice data may be screened to obtain the target voice data, and the target voice data is used in the following process regardless of determining the voice information to be compensated or performing the model training. The above are described from the perspective of model training, and the trained speech synthesis model needs to be applied in a certain service scenario, so the following describes the present invention from the perspective of model application.
Fig. 3 is a schematic flowchart of a voice interaction method in this specification, which specifically includes the following steps:
s301: and responding to a voice interaction request of a user, and determining a voice message sent by the user.
S302: and determining text information replied aiming at the voice message according to the voice message and a preset reply strategy.
S303: and converting the text information into voice synthesis information, inputting the voice synthesis information into a pre-trained voice synthesis model to obtain a synthesized voice message, wherein the voice synthesis information is used for representing voice characteristics corresponding to each voice unit contained in the text information, and the voice synthesis model is obtained by training through a model training method.
S304: and feeding back the synthesized voice message to the user.
In practical applications, a variety of services can be applied to the speech synthesis model in this specification, for example, a speech interaction service, a speech navigation service, and the like, and if a certain text needs to be converted into speech, corresponding speech can be synthesized through the speech synthesis model.
The application process of the speech synthesis model is described below by taking a speech interaction service as an example. The service platform may determine a voice message sent by a user in response to a voice interaction request of the user, and determine, according to the voice message and according to a preset reply strategy, text information replied to the voice message, where the reply strategy may be a preset reply strategy manually set in advance, or the preset reply strategy is text information replied determined by a machine learning model. The voice message refers to the speech of a sentence spoken by the user during the voice interaction.
After the service platform determines the replied text information, the text information needs to be converted into voice and sent to the user, so that the service platform can convert the text information into voice synthesis information, input the voice synthesis information into a pre-trained voice synthesis model to obtain a synthesized voice message, and send the synthesized voice message to the user, so that the user continues to perform voice interaction according to the heard voice message.
The format of the speech synthesis information is consistent with the speech information to be compensated and the compensated speech information, and the speech synthesis information may represent speech features corresponding to each speech unit included in the text information, for example, if a speech unit is a phoneme, the service platform may convert the text information into information on phoneme dimension, and then determine which phonemes should be paused. Moreover, the service platform may further determine which phonemes correspond to the speech that has a certain pronunciation habit feature such as a swallow feature or a lingering feature, so as to determine the speech synthesis information, that is, the speech synthesis information may include phoneme information and pause information, where the phoneme information may not only represent the phonemes but also represent the swallow feature and the lingering feature corresponding to the phonemes.
According to the method, the service platform can perform certain screening steps, so that target voice data with high voice quality can be obtained, the voice data recorded in a specific recording environment does not need to be directly obtained, and the cost can be reduced. Moreover, when model training is carried out, the service platform can compensate the voice information to be compensated which shows the original voice characteristics, so that the voice information after compensation can show the individual pronunciation habit of a speaker in the voice data, the voice synthesis model can be synthesized no matter what pronunciation habit the service platform needs after the model training, and the synthesized voice is more natural after the model training is carried out through the voice data because the voice data can be obtained through a conventional channel (more daily speaking record and the like).
Based on the same idea, the present specification further provides a corresponding apparatus for model training and speech interaction, as shown in fig. 1 or fig. 3.
Fig. 4 is a schematic diagram of a model training apparatus provided in this specification, which specifically includes:
an obtaining module 401, configured to obtain voice data;
an information determining module 402, configured to determine, according to the voice data, to-be-compensated voice information corresponding to the voice data, where the to-be-compensated voice information is used to represent original voice features corresponding to each voice unit included in the voice data;
a feature determining module 403, configured to determine pronunciation habit features corresponding to the voice data, and compensate the voice information to be compensated according to the pronunciation habit features, to obtain compensated voice information;
an input module 404, configured to input the compensated voice information to a voice synthesis model to be trained, so as to obtain a voice synthesis result;
a training module 405, configured to train the speech synthesis model with a goal of minimizing a deviation between the speech synthesis result and the speech data as an optimization goal.
Optionally, the information determining module 402 is specifically configured to segment the voice data to obtain clauses, where each clause includes at least one voice unit; determining the voice information to be compensated corresponding to each clause; and determining the voice information to be compensated corresponding to the voice data according to the voice information to be compensated corresponding to each clause.
Optionally, the information determining module 402 is specifically configured to determine at least one sound source in the voice data, and determine, according to the voice data, audio data corresponding to each sound source in the at least one sound source; selecting a target sound source from the at least one sound source according to the audio data corresponding to each sound source in the at least one sound source; and determining the voice information to be compensated corresponding to the voice data according to the audio data corresponding to the target sound source.
Optionally, the information determining module 402 is specifically configured to determine, according to the audio data corresponding to each sound source in the at least one sound source, an overall voiceprint feature corresponding to the at least one sound source, and determine a sound source voiceprint feature corresponding to each sound source included in the at least one sound source; and selecting a target sound source from the at least one sound source according to the overall voiceprint characteristics and the sound source voiceprint characteristics corresponding to each sound source contained in the at least one sound source.
Optionally, the information determining module 402 is specifically configured to screen out, from the voice data, voice data with voice quality meeting a preset condition as target voice data; and determining the voice information to be compensated corresponding to the voice data according to the target voice data.
Optionally, the information determining module 402 is specifically configured to screen out, from the voice data, voice data with a signal-to-noise ratio not lower than a set signal-to-noise ratio as target voice data; or screening out voice data with the voice amplitude within a set amplitude range from the voice data to serve as target voice data; or screening out voice data with the fundamental frequency within a set fundamental frequency range from the voice data to serve as target voice data; or screening out voice data with the voice time length not less than the set time length from the voice data to serve as target voice data; or screening out the voice data with the speed of speech within the set speed of speech range from the voice data as target voice data.
Optionally, the speech unit includes: a phoneme; the voice information to be compensated comprises at least one of phoneme information and original pause information, and the original pause information is used for representing an original position of a voice pause in the voice data; the pronunciation habit features include: at least one of a pause feature, a drag feature, and a swallow feature.
Optionally, the feature determining module 403 is specifically configured to, for each phoneme included in the speech data, determine that the pronunciation habit feature corresponding to the phoneme is a lingering feature if the duration corresponding to the phoneme exceeds a first set duration, and determine that the pronunciation habit feature corresponding to the phoneme is a swallowing feature if the duration corresponding to the phoneme is less than a second set duration, where the second set duration is shorter than the first set duration.
Fig. 5 is a schematic diagram of a voice interaction apparatus provided in this specification, which specifically includes:
a response module 501, configured to determine, in response to a voice interaction request of a user, a voice message sent by the user;
a text determining module 502, configured to determine, according to the voice message and according to a preset reply policy, text information to be replied to the voice message;
an input module 503, configured to convert the text information into speech synthesis information, and input the speech synthesis information into a pre-trained speech synthesis model to obtain a synthesized speech message;
a feedback module 504, configured to feed back the synthesized voice message to the user, where the voice synthesis information is used to represent voice features corresponding to each voice unit included in the text information, and the voice synthesis model is obtained by training through a model training method.
The present specification also provides a computer-readable storage medium having stored thereon a computer program, the computer program being operable to perform the method of model training and speech interaction described above with reference to fig. 1 or 3.
This specification also provides a schematic block diagram of the electronic device shown in fig. 6. As shown in fig. 6, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, but may also include hardware required for other services. The processor reads a corresponding computer program from the non-volatile memory into the memory and then runs the computer program to implement the method for model training and voice interaction described in fig. 1 or fig. 3. Of course, besides the software implementation, the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.
In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually making an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as abel (advanced Boolean Expression Language), ahdl (alternate Hardware Description Language), traffic, pl (core universal Programming Language), HDCal (jhdware Description Language), lang, Lola, HDL, laspam, hardward Description Language (vhr Description Language), vhal (Hardware Description Language), and vhigh-Language, which are currently used in most common. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.
The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
The above description is only an example of the present specification, and is not intended to limit the present specification. Various modifications and alterations to this description will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.
Claims (13)
1. A method of model training, comprising:
acquiring voice data;
according to the voice data, determining voice information to be compensated corresponding to the voice data, wherein the voice information to be compensated is used for representing original voice characteristics corresponding to all voice units contained in the voice data;
determining pronunciation habit characteristics corresponding to the voice data, and compensating the voice information to be compensated according to the pronunciation habit characteristics to obtain compensated voice information;
inputting the compensated voice information into a voice synthesis model to be trained to obtain a voice synthesis result;
and training the voice synthesis model by taking the minimized deviation between the voice synthesis result and the voice data as an optimization target.
2. The method of claim 1, wherein determining the voice information to be compensated corresponding to the voice data according to the voice data specifically comprises:
segmenting the voice data to obtain clauses, wherein each clause comprises at least one voice unit;
determining the voice information to be compensated corresponding to each clause;
and determining the voice information to be compensated corresponding to the voice data according to the voice information to be compensated corresponding to each clause.
3. The method of claim 1, wherein determining the voice information to be compensated corresponding to the voice data according to the voice data specifically comprises:
determining at least one sound source in the voice data, and determining audio data corresponding to each sound source in the at least one sound source according to the voice data;
selecting a target sound source from the at least one sound source according to the audio data corresponding to each sound source in the at least one sound source;
and determining the voice information to be compensated corresponding to the voice data according to the audio data corresponding to the target sound source.
4. The method of claim 3, wherein selecting a target sound source from the at least one sound source according to the audio data corresponding to each sound source of the at least one sound source comprises:
determining the integral voiceprint characteristics corresponding to the at least one sound source and determining the sound source voiceprint characteristics corresponding to each sound source contained in the at least one sound source according to the audio data corresponding to each sound source in the at least one sound source;
and selecting a target sound source from the at least one sound source according to the overall voiceprint characteristics and the sound source voiceprint characteristics corresponding to each sound source contained in the at least one sound source.
5. The method of claim 1, wherein determining the voice information to be compensated corresponding to the voice data according to the voice data specifically comprises:
screening out voice data with voice quality meeting preset conditions from the voice data to serve as target voice data;
and determining the voice information to be compensated corresponding to the voice data according to the target voice data.
6. The method according to claim 5, wherein the step of screening out, from the voice data, voice data with voice quality meeting a preset condition as target voice data specifically comprises:
screening out voice data with a signal-to-noise ratio not lower than a set signal-to-noise ratio from the voice data as target voice data; or
Screening out voice data with the voice amplitude within a set amplitude range from the voice data to serve as target voice data; or
Screening out voice data with the fundamental frequency within a set fundamental frequency range from the voice data to serve as target voice data; or
Screening out voice data with the voice duration not lower than the set duration from the voice data as target voice data; or
And screening out the voice data with the speed of speech within a set speed of speech range from the voice data as target voice data.
7. The method of claim 1, wherein the speech unit comprises: a phoneme; the voice information to be compensated comprises at least one of phoneme information and original pause information, and the original pause information is used for representing an original position of a voice pause in the voice data; the pronunciation habit features include: at least one of a pause feature, a drag feature, and a swallow feature.
8. The method according to claim 7, wherein determining the pronunciation habit features corresponding to the speech data specifically comprises:
and for each phoneme contained in the voice data, if the duration corresponding to the phoneme exceeds a first set duration, determining that the pronunciation habit feature corresponding to the phoneme is a dragging feature, and if the duration corresponding to the phoneme is less than a second set duration, determining that the pronunciation habit feature corresponding to the phoneme is a swallowing feature, wherein the second set duration is shorter than the first set duration.
9. A method of voice interaction, comprising:
responding to a voice interaction request of a user, and determining a voice message sent by the user;
determining text information replied aiming at the voice message according to the voice message and a preset reply strategy;
converting the text information into voice synthesis information, and inputting the voice synthesis information into a pre-trained voice synthesis model to obtain a synthesized voice message, wherein the voice synthesis information is used for representing voice characteristics corresponding to each voice unit contained in the text information, and the voice synthesis model is obtained by training through the method of any one of claims 1 to 8;
and feeding back the synthesized voice message to the user.
10. An apparatus for model training, comprising:
the acquisition module is used for acquiring voice data;
the information determining module is used for determining to-be-compensated voice information corresponding to the voice data according to the voice data, wherein the to-be-compensated voice information is used for representing original voice characteristics corresponding to each voice unit contained in the voice data;
the characteristic determining module is used for determining pronunciation habit characteristics corresponding to the voice data and compensating the voice information to be compensated according to the pronunciation habit characteristics to obtain compensated voice information;
the input module is used for inputting the compensated voice information into a voice synthesis model to be trained to obtain a voice synthesis result;
and the training module is used for training the voice synthesis model by taking the minimized deviation between the voice synthesis result and the voice data as an optimization target.
11. An apparatus for voice interaction, comprising:
the response module is used for responding to a voice interaction request of a user and determining a voice message sent by the user;
the text determining module is used for determining text information replied aiming at the voice message according to the voice message and a preset reply strategy;
the input module is used for converting the text information into voice synthesis information and inputting the voice synthesis information into a pre-trained voice synthesis model to obtain a synthesized voice message;
a feedback module, configured to feed back the synthesized voice message to the user, where the voice synthesis information is used to represent voice features corresponding to each voice unit included in the text information, and the voice synthesis model is obtained by training according to the method of any one of claims 1 to 8.
12. A computer-readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1-8 or 9.
13. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1 to 8 or 9 when executing the program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110509888.1A CN113314092A (en) | 2021-05-11 | 2021-05-11 | Method and device for model training and voice interaction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110509888.1A CN113314092A (en) | 2021-05-11 | 2021-05-11 | Method and device for model training and voice interaction |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113314092A true CN113314092A (en) | 2021-08-27 |
Family
ID=77372897
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110509888.1A Pending CN113314092A (en) | 2021-05-11 | 2021-05-11 | Method and device for model training and voice interaction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113314092A (en) |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107464554A (en) * | 2017-09-28 | 2017-12-12 | 百度在线网络技术(北京)有限公司 | Phonetic synthesis model generating method and device |
CN109949071A (en) * | 2019-01-31 | 2019-06-28 | 平安科技(深圳)有限公司 | Products Show method, apparatus, equipment and medium based on voice mood analysis |
CN110379407A (en) * | 2019-07-22 | 2019-10-25 | 出门问问(苏州)信息科技有限公司 | Adaptive voice synthetic method, device, readable storage medium storing program for executing and calculating equipment |
CN111402857A (en) * | 2020-05-09 | 2020-07-10 | 广州虎牙科技有限公司 | Speech synthesis model training method and device, electronic equipment and storage medium |
CN111986646A (en) * | 2020-08-17 | 2020-11-24 | 云知声智能科技股份有限公司 | Dialect synthesis method and system based on small corpus |
CN112599113A (en) * | 2020-12-30 | 2021-04-02 | 北京大米科技有限公司 | Dialect voice synthesis method and device, electronic equipment and readable storage medium |
CN112634866A (en) * | 2020-12-24 | 2021-04-09 | 北京猎户星空科技有限公司 | Speech synthesis model training and speech synthesis method, apparatus, device and medium |
-
2021
- 2021-05-11 CN CN202110509888.1A patent/CN113314092A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107464554A (en) * | 2017-09-28 | 2017-12-12 | 百度在线网络技术(北京)有限公司 | Phonetic synthesis model generating method and device |
CN109949071A (en) * | 2019-01-31 | 2019-06-28 | 平安科技(深圳)有限公司 | Products Show method, apparatus, equipment and medium based on voice mood analysis |
CN110379407A (en) * | 2019-07-22 | 2019-10-25 | 出门问问(苏州)信息科技有限公司 | Adaptive voice synthetic method, device, readable storage medium storing program for executing and calculating equipment |
CN111402857A (en) * | 2020-05-09 | 2020-07-10 | 广州虎牙科技有限公司 | Speech synthesis model training method and device, electronic equipment and storage medium |
CN111986646A (en) * | 2020-08-17 | 2020-11-24 | 云知声智能科技股份有限公司 | Dialect synthesis method and system based on small corpus |
CN112634866A (en) * | 2020-12-24 | 2021-04-09 | 北京猎户星空科技有限公司 | Speech synthesis model training and speech synthesis method, apparatus, device and medium |
CN112599113A (en) * | 2020-12-30 | 2021-04-02 | 北京大米科技有限公司 | Dialect voice synthesis method and device, electronic equipment and readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105845125B (en) | Phoneme synthesizing method and speech synthetic device | |
US9218803B2 (en) | Method and system for enhancing a speech database | |
EP1643486B1 (en) | Method and apparatus for preventing speech comprehension by interactive voice response systems | |
CN110599998B (en) | Voice data generation method and device | |
CN109285537B (en) | Acoustic model establishing method, acoustic model establishing device, acoustic model synthesizing method, acoustic model synthesizing device, acoustic model synthesizing equipment and storage medium | |
US10249321B2 (en) | Sound rate modification | |
US11462207B1 (en) | Method and apparatus for editing audio, electronic device and storage medium | |
CN112365878B (en) | Speech synthesis method, device, equipment and computer readable storage medium | |
US7912718B1 (en) | Method and system for enhancing a speech database | |
CN116312471A (en) | Voice migration and voice interaction method and device, electronic equipment and storage medium | |
JP2011233047A (en) | Electronic book device, reproducing method for electronic book and reproducing program for electronic book | |
Stöber et al. | Speech synthesis using multilevel selection and concatenation of units from large speech corpora | |
CN113314092A (en) | Method and device for model training and voice interaction | |
CN114842828A (en) | Volume control method and device for synthesized voice | |
US8510112B1 (en) | Method and system for enhancing a speech database | |
US9570067B2 (en) | Text-to-speech system, text-to-speech method, and computer program product for synthesis modification based upon peculiar expressions | |
CN111429878A (en) | Self-adaptive speech synthesis method and device | |
KR102376552B1 (en) | Voice synthetic apparatus and voice synthetic method | |
US20230169961A1 (en) | Context-aware prosody correction of edited speech | |
CN117219055A (en) | Voice generation method, device, medium and equipment based on tone separation | |
CN116913301A (en) | Voice cloning method and system and electronic equipment | |
CN116403564B (en) | Audio editing method and device, electronic equipment and storage medium | |
JP2007256815A (en) | Voice-reproducing apparatus, voice-reproducing method, and voice reproduction program | |
CN117496941A (en) | Voice data processing method, device and system | |
CN118351826A (en) | Speech synthesis method, device, equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210827 |