WO2021027117A1 - 语音情绪识别方法、装置及计算机可读存储介质 - Google Patents

语音情绪识别方法、装置及计算机可读存储介质 Download PDF

Info

Publication number
WO2021027117A1
WO2021027117A1 PCT/CN2019/116551 CN2019116551W WO2021027117A1 WO 2021027117 A1 WO2021027117 A1 WO 2021027117A1 CN 2019116551 W CN2019116551 W CN 2019116551W WO 2021027117 A1 WO2021027117 A1 WO 2021027117A1
Authority
WO
WIPO (PCT)
Prior art keywords
data set
speech
voice
amplitude
original
Prior art date
Application number
PCT/CN2019/116551
Other languages
English (en)
French (fr)
Inventor
方豪
陆科名
占小杰
王少军
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2021027117A1 publication Critical patent/WO2021027117A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/06Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
    • G10L15/063Training
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/08Speech classification or search
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/08Speech classification or search
    • G10L15/16Speech classification or search using artificial neural networks
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/26Speech to text systems
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02Speech enhancement, e.g. noise reduction or echo cancellation
    • G10L21/0208Noise filtering
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/48Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
    • G10L25/51Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
    • G10L25/63Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination for estimating an emotional state

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a method, device and computer-readable storage medium for receiving voice information input by a user and performing intelligent emotion recognition on the voice information.
  • This application provides a voice emotion recognition method, device, and computer-readable storage medium, the main purpose of which is to receive voice information input by a user and perform intelligent emotion recognition on the voice information.
  • a voice emotion recognition method includes:
  • the user's voice is received, and the user's voice is input into the neural network to obtain an emotion recognition result and output.
  • the present application also provides a voice emotion recognition device, which includes a memory and a processor.
  • the memory stores a voice emotion recognition program that can run on the processor. The following steps are implemented when the recognition program is executed by the processor:
  • the user's voice is received, and the user's voice is input into the neural network to obtain an emotion recognition result and output.
  • the present application also provides a computer-readable storage medium on which a voice emotion recognition program is stored, and the voice emotion recognition program can be executed by one or more processors, To achieve the steps of the voice emotion recognition method as described above.
  • the original speech data set is denoised through the pre-built filter, so the purity of the speech data set is improved.
  • the speech personality classifier is used to classify the speech data set, and the voice loudness under each personality is calculated Amplitude and frequency, as the influence of personality on the amplitude and frequency is increased, so the accuracy of voice emotion recognition is further improved. Therefore, the voice emotion recognition method, device, and computer-readable storage medium proposed in this application can implement accurate and efficient voice emotion recognition functions.
  • FIG. 1 is a schematic flowchart of a voice emotion recognition method provided by an embodiment of this application
  • FIG. 2 is a schematic diagram of the internal structure of a voice emotion recognition device provided by an embodiment of the application.
  • Fig. 3 is a schematic diagram of modules of a voice emotion recognition program in a voice emotion recognition device provided by an embodiment of the application.
  • This application provides a voice emotion recognition method.
  • FIG. 1 it is a schematic flowchart of a voice emotion recognition method provided by an embodiment of this application.
  • the method can be executed by a device, and the device can be implemented by software and/or hardware.
  • the voice emotion recognition method includes:
  • S1 Receive an original speech data set and a tag set, and perform noise reduction processing on the original speech data set based on a pre-built filter to obtain a primary speech data set.
  • the original speech data set is recorded fragments of different people in different scenarios, such as fragments of impassioned speeches by teachers at the college entrance examination mobilization meeting.
  • the label set is divided into two parts, namely the loudness label set and the emotion label set.
  • the loudness label set is a note on the sound loudness of each recording fragment in the original speech data set.
  • the teacher’s impassioned speech fragment the note in the loudness label set is 9.8, where the larger the number, the higher the loudness.
  • the number range of the loudness label set is between [0,10];
  • the emotion label set is a comment on the emotion of each recording segment in the original speech data set, divided into [angry, scared, happy, sad, calm]
  • Five tags, such as the above teacher’s impassioned speech fragment, are noted as happy in the emotional tag set.
  • the noise reduction processing includes inputting the original speech data set to a pre-built filter, and the filter performs a filtering operation on the original speech data set to obtain a speech data output set, and calculating The error between the voice data output set and the original voice data set, if the error is less than the preset threshold A, exit the noise reduction process, if the error is greater than the preset threshold, update the filter And return to the step of performing the filter operation on the original voice data set to obtain the voice data output set, and calculating the error between the voice data output set and the original voice data set, until the error Less than the preset threshold A.
  • M is the voice signal sequence of the original voice data set
  • d is the voice data output set
  • x is the original voice data set
  • w i is the internal weight of each voice signal sequence.
  • u convergence factor the numerical range is ⁇ max represents the maximum value of the wavelength in the original voice data set, Represents the gradient factor of the previous speech signal sequence.
  • the speech personality classifier h ⁇ (x i ) is:
  • x i is each speech segment of the primary speech data set
  • is the adjustment parameter of the speech personality classifier
  • P(y i [0,1]
  • x i , ⁇ ) indicates that the speech segment x i is in ⁇
  • the final loudness range is within the probability value of [0,1]
  • h ⁇ (x i ) is the probability value of all loudness statistics (ie [0,1], [1,2], ..., [9,10])
  • e is the error
  • the primary speech data set and the loudness label set are input into the speech personality classifier, and the speech personality classifier continuously updates the adjustment parameters and judges whether the error e is within the preset threshold B Within the range, until the error e is within the preset threshold B range, the voice personality classifier completes the update of the adjustment parameters, and the personality score set is obtained.
  • the voice personality classifier calculates the probability value of the teacher’s impassioned speech segment x 1 Therefore, it can be seen that the probability that the loudness is [9,10] is 0.98, which is the largest among all 10 intervals, so the final loudness of the teacher's impassioned speech fragment is [9,10].
  • the enhancement calculation is divided into amplitude enhancement calculation and frequency enhancement calculation.
  • the amplitude intensification is calculated as:
  • C is the amplitude of the primary voice data set, that is, the original amplitude
  • C′ is the amplitude of the amplitude enhancement calculation
  • Cmin is the lowest amplitude of the primary voice data set
  • Cmax is the highest amplitude of the primary voice data set
  • R is the personality score set.
  • f is the frequency of the primary voice data set, that is, the original frequency
  • f′ is the frequency of the frequency enhancement calculation
  • fmin is the lowest frequency of the primary voice data set
  • fmax is the highest frequency of the primary voice data set
  • R is the personality score set.
  • the tag set is the emotion tag set
  • the amplitude frequency set is [C′,f′].
  • the neural network includes an input layer, an output layer and a hidden layer.
  • the number of nodes in the input layer is the number of input features, and the amplitude and frequency are determined by the amplitude and frequency set as [C′,f′] as the input of the neural network, so the number of input nodes is 2 .
  • the number of output layer nodes is the number of classification types, and the sentiment label set includes 5 kinds of sentiment discriminant analysis ("angry, scared, happy, sad, peaceful"), namely 5 as the number of output layer nodes .
  • the hidden layer adopts an S-type function hidden layer, and the number of nodes is n:
  • a represents the number of input layer nodes
  • b represents the number of output layer nodes
  • is a constant in [1,10].
  • the training means that the neural network receives the amplitude frequency set and obtains the predicted emotion set, and performs error calculation on the predicted emotion set and the emotion label set to obtain the emotion error value, when the emotion error value After being less than the preset threshold C, the neural network completes training.
  • S5. Receive the user's voice, input the user's voice into the neural network to obtain an emotion recognition result and output it.
  • the receiving method includes receiving the user's real-time voice or a recorded voice segment. For example, if the user receives real-time voice communication with others on the phone, the emotion predicted by the neural network is sad.
  • the invention also provides a voice emotion recognition device.
  • FIG. 2 it is a schematic diagram of the internal structure of a voice emotion recognition device provided by an embodiment of this application.
  • the voice emotion recognition device 1 may be a PC (Personal Computer, personal computer), or a terminal device such as a smart phone, a tablet computer, or a portable computer, or a server.
  • the voice emotion recognition device 1 at least includes a memory 11, a processor 12, a communication bus 13, and a network interface 14.
  • the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc.
  • the memory 11 may be an internal storage unit of the voice emotion recognition device 1 in some embodiments, such as a hard disk of the voice emotion recognition device 1.
  • the memory 11 may also be an external storage device of the voice emotion recognition device 1, such as a plug-in hard disk equipped on the voice emotion recognition device 1, a smart media card (SMC), and a secure digital (Secure Digital). Digital, SD) card, flash card (Flash Card), etc.
  • the memory 11 may also include both an internal storage unit of the voice emotion recognition apparatus 1 and an external storage device.
  • the memory 11 can be used not only to store application software and various data installed in the voice emotion recognition device 1, such as the code of the voice emotion recognition program 01, etc., but also to temporarily store data that has been output or will be output.
  • the processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip, and is used to run the program code or processing stored in the memory 11 Data, such as execution of voice emotion recognition program 01, etc.
  • CPU central processing unit
  • controller microcontroller
  • microprocessor or other data processing chip
  • the communication bus 13 is used to realize the connection and communication between these components.
  • the network interface 14 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface), and is usually used to establish a communication connection between the device 1 and other electronic devices.
  • the device 1 may also include a user interface.
  • the user interface may include a display (Display) and an input unit such as a keyboard (Keyboard).
  • the optional user interface may also include a standard wired interface and a wireless interface.
  • the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light emitting diode) touch device, etc.
  • the display can also be called a display screen or a display unit as appropriate, and is used to display the information processed in the voice emotion recognition device 1 and to display a visualized user interface.
  • Figure 2 only shows the voice emotion recognition device 1 with components 11-14 and the voice emotion recognition program 01. Those skilled in the art can understand that the structure shown in Figure 2 does not constitute a limitation on the voice emotion recognition device 1 It may include fewer or more components than shown, or a combination of some components, or a different component arrangement.
  • the memory 11 stores the voice emotion recognition program 01; when the processor 12 executes the voice emotion recognition program 01 stored in the memory 11, the following steps are implemented:
  • Step 1 Receive an original speech data set and a tag set, and perform noise reduction processing on the original speech data set based on a pre-built filter to obtain a primary speech data set.
  • the original speech data set is recorded fragments of different people in different scenarios, such as fragments of impassioned speeches by teachers at the college entrance examination mobilization meeting.
  • the label set is divided into two parts, namely the loudness label set and the emotion label set.
  • the loudness label set is a note on the sound loudness of each recording fragment in the original speech data set.
  • the teacher’s impassioned speech fragment the note in the loudness label set is 9.8, where the larger the number, the higher the loudness.
  • the number range of the loudness label set is between [0,10];
  • the emotion label set is a comment on the emotion of each recording segment in the original speech data set, divided into [angry, scared, happy, sad, calm]
  • Five tags, such as the above teacher’s impassioned speech fragment, are noted as happy in the emotional tag set.
  • the noise reduction processing includes inputting the original speech data set to a pre-built filter, and the filter performs a filtering operation on the original speech data set to obtain a speech data output set, and calculating The error between the voice data output set and the original voice data set, if the error is less than the preset threshold A, exit the noise reduction process, if the error is greater than the preset threshold, update the filter And return to the step of performing the filter operation on the original voice data set to obtain the voice data output set, and calculating the error between the voice data output set and the original voice data set, until the error Less than the preset threshold A.
  • M is the voice signal sequence of the original voice data set
  • d is the voice data output set
  • x is the original voice data set
  • w i is the internal weight of each voice signal sequence.
  • u convergence factor the numerical range is ⁇ max represents the maximum value of the wavelength in the original voice data set, Represents the gradient factor of the previous speech signal sequence.
  • Step 2 Input the primary speech data set and the label set to the speech personality classifier to obtain a personality score set.
  • the speech personality classifier h ⁇ (x i ) is:
  • x i is each speech segment of the primary speech data set
  • is the adjustment parameter of the speech personality classifier
  • P(y i [0,1]
  • x i , ⁇ ) indicates that the speech segment x i is in ⁇
  • the final loudness range is within the probability value of [0,1]
  • h ⁇ (x i ) is the probability value of all loudness statistics (ie [0,1], [1,2], ..., [9,10])
  • e is the error
  • the primary speech data set and the loudness label set are input into the speech personality classifier, and the speech personality classifier continuously updates the adjustment parameters and judges whether the error e is within the preset threshold B Within the range, until the error e is within the preset threshold B range, the voice personality classifier completes the update of the adjustment parameters, and the personality score set is obtained.
  • the voice personality classifier calculates the probability value of the teacher’s impassioned speech segment x 1 Therefore, it can be seen that the probability that the loudness is [9,10] is 0.98, which is the largest among all 10 intervals, so the final loudness of the teacher's impassioned speech fragment is [9,10].
  • Step 3 Perform enhanced calculation on the amplitude frequency of each voice in the primary speech data set and the personality score set to obtain an amplitude frequency set.
  • the enhancement calculation is divided into amplitude enhancement calculation and frequency enhancement calculation.
  • the amplitude enhancement is calculated as:
  • C is the amplitude of the primary voice data set, that is, the original amplitude
  • C′ is the amplitude of the amplitude enhancement calculation
  • Cmin is the lowest amplitude of the primary voice data set
  • Cmax is the highest amplitude of the primary voice data set
  • R is the personality score set.
  • f is the frequency of the primary voice data set, that is, the original frequency
  • f′ is the frequency of the frequency enhancement calculation
  • fmin is the lowest frequency of the primary voice data set
  • fmax is the highest frequency of the primary voice data set
  • R is the personality score set.
  • Step 4 Input the amplitude frequency set and the label set into a pre-built neural network, and train the neural network.
  • the tag set is the emotion tag set
  • the amplitude frequency set is [C′,f′].
  • the neural network includes an input layer, an output layer and a hidden layer.
  • the number of nodes in the input layer is the number of input features, and the amplitude and frequency are determined by the amplitude and frequency set as [C′,f′] as the input of the neural network, so the number of input nodes is 2 .
  • the number of output layer nodes is the number of classification types, and the sentiment label set includes 5 kinds of sentiment discriminant analysis ("angry, scared, happy, sad, peaceful"), namely 5 as the number of output layer nodes .
  • the hidden layer adopts an S-type function hidden layer, and the number of nodes is n:
  • a represents the number of input layer nodes
  • b represents the number of output layer nodes
  • is a constant in [1,10].
  • the training means that the neural network receives the amplitude frequency set and obtains the predicted emotion set, and performs error calculation on the predicted emotion set and the emotion label set to obtain the emotion error value, when the emotion error value After being less than the preset threshold C, the neural network completes training.
  • Step 5 Receive the user's voice, input the user's voice into the neural network to obtain the emotion recognition result and output it.
  • the receiving method includes receiving the user's real-time voice or a recorded voice segment. For example, if the user receives real-time voice communication with others on the phone, the emotion predicted by the neural network is sad.
  • the voice emotion recognition program can also be divided into one or more modules, and the one or more modules are stored in the memory 11 and run by one or more processors (in this embodiment, The processor 12) is executed to complete the application.
  • the module referred to in the application refers to a series of computer program instruction segments capable of completing specific functions, and is used to describe the execution process of the voice emotion recognition program in the voice emotion recognition device.
  • FIG. 3 it is a schematic diagram of the program modules of the voice emotion recognition program in an embodiment of the voice emotion recognition device of this application.
  • the voice emotion recognition program can be divided into a data receiving and processing module 10 ,
  • the amplitude frequency calculation module 20, the model training module 30, and the emotion recognition output module 40 are exemplary:
  • the data receiving and processing module 10 is configured to receive an original speech data set and a tag set, and perform noise reduction processing on the original speech data set based on a pre-built filter to obtain a primary speech data set.
  • the amplitude frequency calculation module 20 is configured to: input the primary speech data set and the tag set into a speech personality classifier to obtain a personality score set, and compare the amplitude frequency of each voice in the primary speech data set with the personality The score set is enhanced to obtain the amplitude frequency set.
  • the model training module 30 is used for inputting the amplitude frequency set and the label set into a pre-built neural network to train the neural network.
  • the emotion recognition output module 40 is configured to: receive the user's voice, input the user's voice into the neural network to obtain and output the emotion recognition result.
  • an embodiment of the present application also proposes a computer-readable storage medium with a voice emotion recognition program stored on the computer-readable storage medium, and the voice emotion recognition program can be executed by one or more processors to achieve the following operating:
  • the amplitude frequency set and the label set are input into a pre-built neural network, and the neural network is trained.
  • the user's voice is received, and the user's voice is input into the neural network to obtain an emotion recognition result and output.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Computational Linguistics (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Signal Processing (AREA)
  • Artificial Intelligence (AREA)
  • Quality & Reliability (AREA)
  • Child & Adolescent Psychology (AREA)
  • General Health & Medical Sciences (AREA)
  • Hospice & Palliative Care (AREA)
  • Psychiatry (AREA)
  • Evolutionary Computation (AREA)
  • Machine Translation (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种语音情绪识别方法、一种语音情绪识别装置以及一种计算机可读存储介质,语音情绪识别方法包括:接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集(S1),将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集(S2),将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集(S3),将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练(S4),接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出(S5)。

Description

语音情绪识别方法、装置及计算机可读存储介质
本申请基于巴黎公约申明享有2019年8月15日递交的申请号为CN201910768144.4、名称为“语音情绪识别方法、装置及计算机可读存储介质”的中国专利申请的优先权,该中国专利申请的整体内容以参考的方式结合在本申请中。
技术领域
本申请涉及人工智能技术领域,尤其涉及一种接收用户输入语音信息,对所述语音信息进行智能情绪识别的方法、装置及计算机可读存储介质。
背景技术
由于每个人性格、说话风格、控制情绪表达的能力等不同,情绪变化在语言上的表现程度也就不同,而目前的语音情绪识别模型都是通用模型,无法根据不同人的不同说话特点进行个性化地判别,导致出现很多的错判漏判。再者,通用模型的分类精度有限,导致很多差异不大的情绪也无法区分。
发明内容
本申请提供一种语音情绪识别方法、装置及计算机可读存储介质,其主要目的在于接收用户输入的语音信息,对所述语音信息进行智能情绪识别。
为实现上述目的,本申请提供的一种语音情绪识别方法,包括:
接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集;
将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集;
将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集;
将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练;
接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出。
此外,为实现上述目的,本申请还提供一种语音情绪识别装置,该装置包括存储器和处理器,所述存储器中存储有可在所述处理器上运行的语音情绪识别程序,所述语音情绪识别程序被所述处理器执行时实现如下步骤:
接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集;
将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集;
将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集;
将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练;
接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出。
此外,为实现上述目的,本申请还提供一种计算机可读存储介质,所述计算机可读存储介质上存储有语音情绪识别程序,所述语音情绪识别程序可被一个或者多个处理器执行,以实现如上所述的语音情绪识别方法的步骤。
本申请由于通过预先构建的滤波器对原始语音数据集进行降噪处理,故提高了语音数据集的纯洁度,另外利用语音性格分类器对语音数据集进行性格分类,通过各个性格下声音响度计算幅度和频率,由于增加了性格对幅度和频率的影响,所以进一步提高了语音情绪识别的准确性。因此本申请提出的语音情绪识别方法、装置及计算机可读存储介质可以实现精准高效的语音情绪识别功能。
附图说明
图1为本申请一实施例提供的语音情绪识别方法的流程示意图;
图2为本申请一实施例提供的语音情绪识别装置的内部结构示意图;
图3为本申请一实施例提供的语音情绪识别装置中语音情绪识别程序的 模块示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本申请提供一种语音情绪识别方法。参照图1所示,为本申请一实施例提供的语音情绪识别方法的流程示意图。该方法可以由一个装置执行,该装置可以由软件和/或硬件实现。
在本实施例中,语音情绪识别方法包括:
S1、接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集。
较佳地,所述原始语音数据集是不同人在不同场景下的录音片段,如高考动员大会上,老师慷慨激昂的演讲片段等。
所述标签集分为两个部分,分别为响度标签集和情绪标签集。所述响度标签集是对所述原始语音数据集中各个录音片段声音响度的备注,如所述老师慷慨激昂的演讲片段,在所述响度标签集的备注为9.8,其中数字越大代表响度越高,所述响度标签集的数字范围在[0,10]之间;所述情绪标签集是对所述原始语音数据集中各个录音片段情绪的备注,分为[生气、害怕、高兴、伤心、平静]五个标签,如上述老师慷慨激昂的演讲片段在所述情绪标签集备注为高兴。
较佳地,由于现实生活中语音信号一般都带有噪声,在进行情绪识别前需要对语音信号进行降噪处理。本申请较佳实施例中,所述降噪处理包括将所述原始语音数据集输入至预先构建的滤波器,所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集,计算所述语音数据输出集与所述原始语音数据集的误差,若所述误差小于预设阈值A时退出所述降噪处理,若所述误差大于所述预设阈值时,更新所述滤波器的内部权重,并返回执行所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集,计算所述语音数据输出集与所述原始语音数据集的误差的步骤,直至所述误差小 于预设阈值A。
进一步地,所述误差e 1通过下述方法结算:
Figure PCTCN2019116551-appb-000001
其中,M为所述原始语音数据集的语音信号序列,d为所述语音数据输出集,x为所述原始语音数据集,w i为所述各个语音信号序列的内部权重。
进一步地,所述w i为:
Figure PCTCN2019116551-appb-000002
其中,u收敛因子,数值范围在
Figure PCTCN2019116551-appb-000003
λ max表示所述原始语音数据集中波长的最大值,
Figure PCTCN2019116551-appb-000004
表示前一语音信号序列的梯度因子。
S2、将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集。
所述语音性格分类器h θ(x i)为:
Figure PCTCN2019116551-appb-000005
其中,x i为所述初级语音数据集各个语音片段,θ为所述语音性格分类器的调节参数,P(y i=[0,1]|x i,θ)表示语音片段x i在θ为调节参数的前提下,最后的响度范围在[0,1]的概率值,而h θ(x i)是统计出所有响度的概率值(即[0,1],[1,2],…,[9,10]),e为误差,
Figure PCTCN2019116551-appb-000006
表示各个响度下所述调节参数的转置矩阵。
较佳地,将所述初级语音数据集及所述响度标签集输入至所述语音性格分类器中,所述语音性格分类器不断更新所述调节参数,并判断误差e是否在预设阈值B范围内,直至满足所述误差e在所述预设阈值B范围内,所述语音性格分类器更新所述调节参数完成,得到所述性格分值集。如所述语音性格分类器计算所述老师慷慨激昂的演讲片段x 1的概率值
Figure PCTCN2019116551-appb-000007
因此可看出响度为[9,10]的概率为0.98,在所有10个区间中最大,因此所述老师慷慨激昂的演讲片段最终的响度为[9,10]。
S3、将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集。
较佳地,所述强化计算分为幅度强化计算和频率强化计算。所述幅度强 化计算为:
Figure PCTCN2019116551-appb-000008
其中,C为所述初级语音数据集的幅度,即原幅度,C′为所述幅度强化计算的幅度,Cmin所述初级语音数据集的最低幅度,Cmax为所述初级语音数据集的最高幅度,r为所述性格分值集。
进一步地,所述频率强化计算为:
Figure PCTCN2019116551-appb-000009
其中,f为所述初级语音数据集的频率,即原频率,f′为所述频率强化计算的频率,fmin所述初级语音数据集的最低频率,fmax为所述初级语音数据集的最高频率,r为所述性格分值集。
S4、将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练。
较佳地,所述标签集为所述情绪标签集,所述幅度频率集为[C′,f′]。
所述神经网络包括输入层、输出层以及隐含层。所述输入层的节点数即为输入特征的个数,由所述幅度频率集为[C′,f′]确定了幅值、频率作为所述神经网络的输入,因此输入节点的数目为2。所述输出层节点数目即为分类类型数,由所述情绪标签集是包括5种的情绪判别分析(“生气、害怕、高兴、伤心、平静”),即5作为所述输出层节点的数目。所述隐含层采用S型函数隐含层,其节点数n:
Figure PCTCN2019116551-appb-000010
其中,a表示输入层节点的数目,b表示输出层节点的数目,ɑ为[1,10]内的常数。
进一步地,所述训练是指所述神经网络接收所述幅度频率集并得到预测情绪集,将所述预测情绪集与所述情绪标签集进行误差计算得到情绪误差值,当所述情绪误差值小于预设阈值C后,所述神经网络完成训练。
S5、接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出。
优选地,所述接收方式包括接收用户实时的声音或已经录制好的语音片段等。如接收用户实时的和别人的电话交流声音,经过所述神经网络预测得 到的情绪为伤心。
发明还提供一种语音情绪识别装置。参照图2所示,为本申请一实施例提供的语音情绪识别装置的内部结构示意图。
在本实施例中,所述语音情绪识别装置1可以是PC(Personal Computer,个人电脑),或者是智能手机、平板电脑、便携计算机等终端设备,也可以是一种服务器等。该语音情绪识别装置1至少包括存储器11、处理器12,通信总线13,以及网络接口14。
其中,存储器11至少包括一种类型的可读存储介质,所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如,SD或DX存储器等)、磁性存储器、磁盘、光盘等。存储器11在一些实施例中可以是语音情绪识别装置1的内部存储单元,例如该语音情绪识别装置1的硬盘。存储器11在另一些实施例中也可以是语音情绪识别装置1的外部存储设备,例如语音情绪识别装置1上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,存储器11还可以既包括语音情绪识别装置1的内部存储单元也包括外部存储设备。存储器11不仅可以用于存储安装于语音情绪识别装置1的应用软件及各类数据,例如语音情绪识别程序01的代码等,还可以用于暂时地存储已经输出或者将要输出的数据。
处理器12在一些实施例中可以是一中央处理器(Central Processing Unit,CPU)、控制器、微控制器、微处理器或其他数据处理芯片,用于运行存储器11中存储的程序代码或处理数据,例如执行语音情绪识别程序01等。
通信总线13用于实现这些组件之间的连接通信。
网络接口14可选的可以包括标准的有线接口、无线接口(如WI-FI接口),通常用于在该装置1与其他电子设备之间建立通信连接。
可选地,该装置1还可以包括用户接口,用户接口可以包括显示器(Display)、输入单元比如键盘(Keyboard),可选的用户接口还可以包括标准的有线接口、无线接口。可选地,在一些实施例中,显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode,有机发光二极管)触摸器等。其中,显示器也可以适当的称为显示屏 或显示单元,用于显示在语音情绪识别装置1中处理的信息以及用于显示可视化的用户界面。
图2仅示出了具有组件11-14以及语音情绪识别程序01的语音情绪识别装置1,本领域技术人员可以理解的是,图2示出的结构并不构成对语音情绪识别装置1的限定,可以包括比图示更少或者更多的部件,或者组合某些部件,或者不同的部件布置。
在图2所示的装置1实施例中,存储器11中存储有语音情绪识别程序01;处理器12执行存储器11中存储的语音情绪识别程序01时实现如下步骤:
步骤一、接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集。
较佳地,所述原始语音数据集是不同人在不同场景下的录音片段,如高考动员大会上,老师慷慨激昂的演讲片段等。
所述标签集分为两个部分,分别为响度标签集和情绪标签集。所述响度标签集是对所述原始语音数据集中各个录音片段声音响度的备注,如所述老师慷慨激昂的演讲片段,在所述响度标签集的备注为9.8,其中数字越大代表响度越高,所述响度标签集的数字范围在[0,10]之间;所述情绪标签集是对所述原始语音数据集中各个录音片段情绪的备注,分为[生气、害怕、高兴、伤心、平静]五个标签,如上述老师慷慨激昂的演讲片段在所述情绪标签集备注为高兴。
较佳地,由于现实生活中语音信号一般都带有噪声,在进行情绪识别前需要对语音信号进行降噪处理。本申请较佳实施例中,所述降噪处理包括将所述原始语音数据集输入至预先构建的滤波器,所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集,计算所述语音数据输出集与所述原始语音数据集的误差,若所述误差小于预设阈值A时退出所述降噪处理,若所述误差大于所述预设阈值时,更新所述滤波器的内部权重,并返回执行所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集,计算所述语音数据输出集与所述原始语音数据集的误差的步骤,直至所述误差小于预设阈值A。
进一步地,所述误差e 1通过下述方法结算:
Figure PCTCN2019116551-appb-000011
其中,M为所述原始语音数据集的语音信号序列,d为所述语音数据输出集,x为所述原始语音数据集,w i为所述各个语音信号序列的内部权重。
进一步地,所述w i为:
Figure PCTCN2019116551-appb-000012
其中,u收敛因子,数值范围在
Figure PCTCN2019116551-appb-000013
λ max表示所述原始语音数据集中波长的最大值,
Figure PCTCN2019116551-appb-000014
表示前一语音信号序列的梯度因子。
步骤二、将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集。
所述语音性格分类器h θ(x i)为:
Figure PCTCN2019116551-appb-000015
其中,x i为所述初级语音数据集各个语音片段,θ为所述语音性格分类器的调节参数,P(y i=[0,1]|x i,θ)表示语音片段x i在θ为调节参数的前提下,最后的响度范围在[0,1]的概率值,而h θ(x i)是统计出所有响度的概率值(即[0,1],[1,2],…,[9,10]),e为误差,
Figure PCTCN2019116551-appb-000016
表示各个响度下所述调节参数的转置矩阵。
较佳地,将所述初级语音数据集及所述响度标签集输入至所述语音性格分类器中,所述语音性格分类器不断更新所述调节参数,并判断误差e是否在预设阈值B范围内,直至满足所述误差e在所述预设阈值B范围内,所述语音性格分类器更新所述调节参数完成,得到所述性格分值集。如所述语音性格分类器计算所述老师慷慨激昂的演讲片段x 1的概率值
Figure PCTCN2019116551-appb-000017
因此可看出响度为[9,10]的概率为0.98,在所有10个区间中最大,因此所述老师慷慨激昂的演讲片段最终的响度为[9,10]。
步骤三、将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集。
较佳地,所述强化计算分为幅度强化计算和频率强化计算。所述幅度强化计算为:
Figure PCTCN2019116551-appb-000018
其中,C为所述初级语音数据集的幅度,即原幅度,C′为所述幅度强化计算的幅度,Cmin所述初级语音数据集的最低幅度,Cmax为所述初级语音数据集的最高幅度,r为所述性格分值集。
进一步地,所述频率强化计算为:
Figure PCTCN2019116551-appb-000019
其中,f为所述初级语音数据集的频率,即原频率,f′为所述频率强化计算的频率,fmin所述初级语音数据集的最低频率,fmax为所述初级语音数据集的最高频率,r为所述性格分值集。
步骤四、将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练。
较佳地,所述标签集为所述情绪标签集,所述幅度频率集为[C′,f′]。
所述神经网络包括输入层、输出层以及隐含层。所述输入层的节点数即为输入特征的个数,由所述幅度频率集为[C′,f′]确定了幅值、频率作为所述神经网络的输入,因此输入节点的数目为2。所述输出层节点数目即为分类类型数,由所述情绪标签集是包括5种的情绪判别分析(“生气、害怕、高兴、伤心、平静”),即5作为所述输出层节点的数目。所述隐含层采用S型函数隐含层,其节点数n:
Figure PCTCN2019116551-appb-000020
其中,a表示输入层节点的数目,b表示输出层节点的数目,ɑ为[1,10]内的常数。
进一步地,所述训练是指所述神经网络接收所述幅度频率集并得到预测情绪集,将所述预测情绪集与所述情绪标签集进行误差计算得到情绪误差值,当所述情绪误差值小于预设阈值C后,所述神经网络完成训练。
步骤五、接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出。
优选地,所述接收方式包括接收用户实时的声音或已经录制好的语音片段等。如接收用户实时的和别人的电话交流声音,经过所述神经网络预测得到的情绪为伤心。
可选地,在其他实施例中,语音情绪识别程序还可以被分割为一个或者 多个模块,一个或者多个模块被存储于存储器11中,并由一个或多个处理器(本实施例为处理器12)所执行以完成本申请,本申请所称的模块是指能够完成特定功能的一系列计算机程序指令段,用于描述语音情绪识别程序在语音情绪识别装置中的执行过程。
例如,参照图3所示,为本申请语音情绪识别装置一实施例中的语音情绪识别程序的程序模块示意图,该实施例中,所述语音情绪识别程序可以被分割为数据接收及处理模块10、幅度频率计算模块20、模型训练模块30、情绪识别输出模块40示例性地:
所述数据接收及处理模块10用于:接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集。
所述幅度频率计算模块20用于:将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集,将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集。
所述模型训练模块30用于:将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练。
所述情绪识别输出模块40用于:接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出。
上述数据接收及处理模块10、幅度频率计算模块20、模型训练模块30、情绪识别输出模块40等程序模块被执行时所实现的功能或操作步骤与上述实施例大体相同,在此不再赘述。
此外,本申请实施例还提出一种计算机可读存储介质,所述计算机可读存储介质上存储有语音情绪识别程序,所述语音情绪识别程序可被一个或多个处理器执行,以实现如下操作:
接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集。
将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集,将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集。
将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练。
接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出。
需要说明的是,上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。并且本文中的术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、装置、物品或者方法不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、装置、物品或者方法所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种语音情绪识别方法,其特征在于,所述方法包括:
    接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集;
    将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集;
    将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集;
    将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练;
    接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出。
  2. 如权利要求1所述的语音情绪识别方法,其特征在于,所述降噪处理,包括:
    将所述原始语音数据集输入至预先构建的滤波器;
    利用所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集;
    计算所述语音数据输出集与所述原始语音数据集的误差;
    在所述误差大于预设阈值A时更新所述滤波器的内部权重,并返回执行利用所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集及计算所述语音数据输出集与所述原始语音数据集的误差,直至所述误差小于所述预设阈值A时完成所述降噪处理。
  3. 如权利要求2所述的语音情绪识别方法,其特征在于,所述误差通过下述方法计算:
    Figure PCTCN2019116551-appb-100001
    其中,e 1为所述误差,M为所述原始语音数据集的语音信号序列,d为所述语音数据输出集,x为所述原始语音数据集,w i为所述各个语音信号序列的内部权重,所述w i通过下述方法计算:
    Figure PCTCN2019116551-appb-100002
    其中,u收敛因子,数值范围在
    Figure PCTCN2019116551-appb-100003
    λ max表示所述原始语音数据集中波长的最大值,
    Figure PCTCN2019116551-appb-100004
    表示前一语音信号序列的梯度因子。
  4. 如权利要求1至3中任意一项所述的语音情绪识别方法,其特征在于,所述语音性格分类器为:
    Figure PCTCN2019116551-appb-100005
    其中h θ(x i)为所述语音性格分类器,x i为所述初级语音数据集各个语音片段,θ为所述语音性格分类器的调节参数,P(y i=[0,1]|x i,θ)表示语音片段x i在θ为调节参数的前提下,最后的响度范围在[0,1]的概率值,而h θ(x i)是统计出所有响度的概率值,e为误差,
    Figure PCTCN2019116551-appb-100006
    表示各个响度下所述调节参数的转置矩阵。
  5. 如权利要求1所述的语音情绪识别方法,其特征在于,所述强化计算分为幅度强化计算和频率强化计算;
    其中,所述幅度强化计算为:
    Figure PCTCN2019116551-appb-100007
    其中,C为所述初级语音数据集的幅度,即原幅度,C′为所述幅度强化计算的幅度,Cmin所述初级语音数据集的最低幅度,Cmax为所述初级语音数据集的最高幅度,r为所述性格分值集;
    所述频率强化计算为:
    Figure PCTCN2019116551-appb-100008
    其中,f为所述初级语音数据集的频率,即原频率,f′为所述频率强化计算的频率,fmin所述初级语音数据集的最低频率,fmax为所述初级语音数据集的最高频率,r为所述性格分值集。
  6. 如权利要求5所述的语音情绪识别方法,其特征在于,所述标签集为情绪标签集,所述幅度频率集为[C′,f′],所述神经网络包括输入层、输出层以及隐含层。
  7. 如权利要求6所述的语音情绪识别方法,其特征在于,所述隐含层采用S型函数隐含层,其节点数n:
    Figure PCTCN2019116551-appb-100009
    其中,a表示输入层节点的数目,b表示输出层节点的数目,ɑ为[1,10]内的常数。
  8. 一种语音情绪识别装置,其特征在于,所述装置包括存储器和处理器,所述存储器上存储有可在所述处理器上运行的语音情绪识别程序,所述语音情绪识别程序被所述处理器执行时实现如下步骤:
    接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集;
    将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集;
    将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集;
    将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练;
    接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出。
  9. 如权利要求8所述的语音情绪识别装置,其特征在于,所述降噪处理,包括:
    将所述原始语音数据集输入至预先构建的滤波器;
    利用所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集;
    计算所述语音数据输出集与所述原始语音数据集的误差;
    在所述误差大于预设阈值A时更新所述滤波器的内部权重,并返回执行利用所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集及计算所述语音数据输出集与所述原始语音数据集的误差,直至所述误差小于所述预设阈值A时完成所述降噪处理。
  10. 如权利要求9所述的语音情绪识别装置,其特征在于,所述误差通过下述方法计算:
    Figure PCTCN2019116551-appb-100010
    其中,e 1为所述误差,M为所述原始语音数据集的语音信号序列,d为所述语音数据输出集,x为所述原始语音数据集,w i为所述各个语音信号序列的内部权重,所述w i通过下述方法计算:
    Figure PCTCN2019116551-appb-100011
    其中,u收敛因子,数值范围在
    Figure PCTCN2019116551-appb-100012
    λ max表示所述原始语音数据集中波长的最大值,
    Figure PCTCN2019116551-appb-100013
    表示前一语音信号序列的梯度因子。
  11. 如权利要求8-10中任意一项所述的语音情绪识别装置,其特征在于,所述语音性格分类器为:
    Figure PCTCN2019116551-appb-100014
    其中h θ(x i)为所述语音性格分类器,x i为所述初级语音数据集各个语音片段,θ为所述语音性格分类器的调节参数,P(y i=[0,1]|x i,θ)表示语音片段x i在θ为调节参数的前提下,最后的响度范围在[0,1]的概率值,而h θ(x i)是统计出所有响度的概率值,e为误差,
    Figure PCTCN2019116551-appb-100015
    表示各个响度下所述调节参数的转置矩阵。
  12. 如权利要求8所述的语音情绪识别装置,其特征在于,所述强化计算分为幅度强化计算和频率强化计算;
    其中,所述幅度强化计算为:
    Figure PCTCN2019116551-appb-100016
    其中,C为所述初级语音数据集的幅度,即原幅度,C′为所述幅度强化计算的幅度,Cmin所述初级语音数据集的最低幅度,Cmax为所述初级语音数据集的最高幅度,r为所述性格分值集;
    所述频率强化计算为:
    Figure PCTCN2019116551-appb-100017
    其中,f为所述初级语音数据集的频率,即原频率,f′为所述频率强化计算的频率,fmin所述初级语音数据集的最低频率,fmax为所述初级语音数据集的最高频率,r为所述性格分值集。
  13. 如权利要求12所述的语音情绪识别装置,其特征在于,所述标签集为情绪标签集,所述幅度频率集为[C′,f′],所述神经网络包括输入层、输出层 以及隐含层。
  14. 如权利要求13所述的语音情绪识别装置,其特征在于,所述隐含层采用S型函数隐含层,其节点数n:
    Figure PCTCN2019116551-appb-100018
    其中,a表示输入层节点的数目,b表示输出层节点的数目,ɑ为[1,10]内的常数。
  15. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质上存储有语音情绪识别程序,所述语音情绪识别程序可被一个或者多个处理器执行时,实现如下步骤:
    接收原始语音数据集及标签集,基于预先构建的滤波器对所述原始语音数据集进行降噪处理得到初级语音数据集;
    将所述初级语音数据集及所述标签集输入至语音性格分类器得到性格分值集;
    将所述初级语音数据集中各语音的幅度频率与所述性格分值集进行强化计算得到幅度频率集;
    将所述幅度频率集及标签集输入预先构建的神经网络中,对所述神经网络进行训练;
    接收用户的语音,将所述用户的语音输入至所述神经网络中得到情绪识别结果并输出。
  16. 如权利要求15所述的计算机可读存储介质,其特征在于,所述降噪处理,包括:
    将所述原始语音数据集输入至预先构建的滤波器;
    利用所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集;
    计算所述语音数据输出集与所述原始语音数据集的误差;
    在所述误差大于预设阈值A时更新所述滤波器的内部权重,并返回执行利用所述滤波器对所述原始语音数据集进行滤波操作得到语音数据输出集及计算所述语音数据输出集与所述原始语音数据集的误差,直至所述误差小于所述预设阈值A时完成所述降噪处理。
  17. 如权利要求16所述的计算机可读存储介质,其特征在于,所述误差 通过下述方法计算:
    Figure PCTCN2019116551-appb-100019
    其中,e 1为所述误差,M为所述原始语音数据集的语音信号序列,d为所述语音数据输出集,x为所述原始语音数据集,w i为所述各个语音信号序列的内部权重,所述w i通过下述方法计算:
    Figure PCTCN2019116551-appb-100020
    其中,u收敛因子,数值范围在
    Figure PCTCN2019116551-appb-100021
    λ max表示所述原始语音数据集中波长的最大值,
    Figure PCTCN2019116551-appb-100022
    表示前一语音信号序列的梯度因子。
  18. 如权利要求15-17中任一项所述的计算机可读存储介质,其特征在于,所述语音性格分类器为:
    Figure PCTCN2019116551-appb-100023
    其中h θ(x i)为所述语音性格分类器,x i为所述初级语音数据集各个语音片段,θ为所述语音性格分类器的调节参数,P(y i=[0,1]|x i,θ)表示语音片段x i在θ为调节参数的前提下,最后的响度范围在[0,1]的概率值,而h θ(x i)是统计出所有响度的概率值,e为误差,
    Figure PCTCN2019116551-appb-100024
    表示各个响度下所述调节参数的转置矩阵。
  19. 如权利要求15所述的计算机可读存储介质,其特征在于,所述强化计算分为幅度强化计算和频率强化计算;
    其中,所述幅度强化计算为:
    Figure PCTCN2019116551-appb-100025
    其中,C为所述初级语音数据集的幅度,即原幅度,C′为所述幅度强化计算的幅度,Cmin所述初级语音数据集的最低幅度,Cmax为所述初级语音数据集的最高幅度,r为所述性格分值集;
    所述频率强化计算为:
    Figure PCTCN2019116551-appb-100026
    其中,f为所述初级语音数据集的频率,即原频率,f′为所述频率强化计算的频率,fmin所述初级语音数据集的最低频率,fmax为所述初级语音数据集的最高频率,r为所述性格分值集。
  20. 如权利要求19所述的计算机可读存储介质,其特征在于,所述标签集为情绪标签集,所述幅度频率集为[C′,f′],所述神经网络包括输入层、输出层以及隐含层。
PCT/CN2019/116551 2019-08-15 2019-11-08 语音情绪识别方法、装置及计算机可读存储介质 WO2021027117A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910768144.4A CN110570844B (zh) 2019-08-15 2019-08-15 语音情绪识别方法、装置及计算机可读存储介质
CN201910768144.4 2019-08-15

Publications (1)

Publication Number Publication Date
WO2021027117A1 true WO2021027117A1 (zh) 2021-02-18

Family

ID=68775373

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/116551 WO2021027117A1 (zh) 2019-08-15 2019-11-08 语音情绪识别方法、装置及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN110570844B (zh)
WO (1) WO2021027117A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112818841A (zh) * 2021-01-29 2021-05-18 北京搜狗科技发展有限公司 一种识别用户情绪的方法及相关装置
CN113651199A (zh) * 2021-08-26 2021-11-16 江苏社家通信息科技有限公司 手机语音通过iccid码或imei的电梯控制方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101685634A (zh) * 2008-09-27 2010-03-31 上海盛淘智能科技有限公司 一种儿童语音情感识别方法
CN103634472A (zh) * 2013-12-06 2014-03-12 惠州Tcl移动通信有限公司 根据通话语音判断用户心情及性格的方法、***及手机
CN108735232A (zh) * 2017-04-24 2018-11-02 北京理工大学 一种性格识别方法和装置
CN109473122A (zh) * 2018-11-12 2019-03-15 平安科技(深圳)有限公司 基于检测模型的情绪分析方法、装置及终端设备
CN109767787A (zh) * 2019-01-28 2019-05-17 腾讯科技(深圳)有限公司 情绪识别方法、设备及可读存储介质
US20190213400A1 (en) * 2018-01-05 2019-07-11 Samsung Electronics Co., Ltd. Method and apparatus with emotion recognition
CN110059211A (zh) * 2019-03-28 2019-07-26 华为技术有限公司 记录用户情感的方法及相关装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180082679A1 (en) * 2016-09-18 2018-03-22 Newvoicemedia, Ltd. Optimal human-machine conversations using emotion-enhanced natural speech using hierarchical neural networks and reinforcement learning
CN106598948B (zh) * 2016-12-19 2019-05-03 杭州语忆科技有限公司 基于长短期记忆神经网络结合自动编码器的情绪识别方法
CN107657964B (zh) * 2017-08-15 2021-01-05 西北大学 基于声学特征和稀疏数学的抑郁症辅助检测方法及分类器
CN109272993A (zh) * 2018-08-21 2019-01-25 中国平安人寿保险股份有限公司 语音类别的识别方法、装置、计算机设备和存储介质
CN109389992A (zh) * 2018-10-18 2019-02-26 天津大学 一种基于振幅和相位信息的语音情感识别方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101685634A (zh) * 2008-09-27 2010-03-31 上海盛淘智能科技有限公司 一种儿童语音情感识别方法
CN103634472A (zh) * 2013-12-06 2014-03-12 惠州Tcl移动通信有限公司 根据通话语音判断用户心情及性格的方法、***及手机
CN108735232A (zh) * 2017-04-24 2018-11-02 北京理工大学 一种性格识别方法和装置
US20190213400A1 (en) * 2018-01-05 2019-07-11 Samsung Electronics Co., Ltd. Method and apparatus with emotion recognition
CN109473122A (zh) * 2018-11-12 2019-03-15 平安科技(深圳)有限公司 基于检测模型的情绪分析方法、装置及终端设备
CN109767787A (zh) * 2019-01-28 2019-05-17 腾讯科技(深圳)有限公司 情绪识别方法、设备及可读存储介质
CN110059211A (zh) * 2019-03-28 2019-07-26 华为技术有限公司 记录用户情感的方法及相关装置

Also Published As

Publication number Publication date
CN110570844B (zh) 2023-05-05
CN110570844A (zh) 2019-12-13

Similar Documents

Publication Publication Date Title
WO2020224213A1 (zh) 语句意图识别方法、装置及计算机可读存储介质
US11010645B2 (en) Interactive artificial intelligence analytical system
CN111368996B (zh) 可传递自然语言表示的重新训练投影网络
WO2021233112A1 (zh) 基于多模态机器学习的翻译方法、装置、设备及存储介质
WO2020232860A1 (zh) 语音合成方法、装置及计算机可读存储介质
WO2021208287A1 (zh) 用于情绪识别的语音端点检测方法、装置、电子设备及存储介质
US20210201143A1 (en) Computing device and method of classifying category of data
CN107783960B (zh) 用于抽取信息的方法、装置和设备
Sahu et al. Recurrent neural network models for disease name recognition using domain invariant features
CN111985240B (zh) 命名实体识别模型的训练方法、命名实体识别方法及装置
WO2021051577A1 (zh) 语音情绪识别方法、装置、设备及存储介质
WO2020238046A1 (zh) 人声智能检测方法、装置及计算机可读存储介质
CN111951805A (zh) 一种文本数据处理方法及装置
WO2020261234A1 (en) System and method for sequence labeling using hierarchical capsule based neural network
CN111833845B (zh) 多语种语音识别模型训练方法、装置、设备及存储介质
CN113094578B (zh) 基于深度学习的内容推荐方法、装置、设备及存储介质
CN112395979B (zh) 基于图像的健康状态识别方法、装置、设备及存储介质
CN110502610A (zh) 基于文本语义相似度的智能语音签名方法、装置及介质
CN112418059A (zh) 一种情绪识别的方法、装置、计算机设备及存储介质
WO2021027117A1 (zh) 语音情绪识别方法、装置及计算机可读存储介质
US20230260536A1 (en) Interactive artificial intelligence analytical system
WO2023030105A1 (zh) 训练自然语言处理模型和自然语言处理的方法、电子设备
CN117296058A (zh) 作为简单有效的对抗攻击方法的变体不一致攻击(via)
JP2023088281A (ja) 倫理的視点に基づいたイメージ生成
WO2022073341A1 (zh) 基于语音语义的疾病实体匹配方法、装置及计算机设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19941474

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19941474

Country of ref document: EP

Kind code of ref document: A1