CN106782550A - A kind of automatic speech recognition system based on dsp chip - Google Patents
A kind of automatic speech recognition system based on dsp chip Download PDFInfo
- Publication number
- CN106782550A CN106782550A CN201611064684.7A CN201611064684A CN106782550A CN 106782550 A CN106782550 A CN 106782550A CN 201611064684 A CN201611064684 A CN 201611064684A CN 106782550 A CN106782550 A CN 106782550A
- Authority
- CN
- China
- Prior art keywords
- module
- speech recognition
- unit
- voice signal
- pattern matching
- 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
- 238000013528 artificial neural network Methods 0.000 claims abstract description 29
- 238000007781 pre-processing Methods 0.000 claims abstract description 19
- 238000001514 detection method Methods 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 230000005236 sound signal Effects 0.000 claims 1
- 238000000034 method Methods 0.000 abstract description 8
- 238000012545 processing Methods 0.000 abstract description 4
- 238000004458 analytical method Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000007935 neutral effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000857 drug effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
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
- G10L15/00—Speech recognition
- G10L15/26—Speech to text systems
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)
- Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
Abstract
The invention discloses a kind of automatic speech recognition system based on dsp chip, including voice signal acquisition device, wavelet filter, speech signal pre-processing module, speech recognition module, neural network module, Pattern Matching Module, speech recognition output module and dsp chip, the voice signal acquisition device, wavelet filter, speech signal pre-processing module, speech recognition module, Pattern Matching Module and speech recognition output module are sequentially connected, and speech recognition module and Pattern Matching Module are connected with neural network module.The present invention is based on Speech processing, small echo and neural network theory and method, have studied the Dynamic Recognition of voice signal, and small echo and neural network theory are applied into speech recognition with method, automatically voice can be identified, simple structure, easy to use, low cost.
Description
Technical field
The present invention relates to technical field of voice recognition, more particularly to a kind of automatic speech recognition system based on dsp chip.
Background technology
Automatic speech recognition is always the ideal that the mankind pursue, and is also the direction that immediate and mid-term scientific research personnel seek assiduously；
Its final goal is to allow machine to understand the language of the mankind, and performs corresponding function；Although over 50 years, people is in field of speech recognition
Considerable progress is achieved, but we can be clearly seen that, also have very big gap apart from preferable target；With computer
Fast development, speech recognition develops into a poor interdisciplinary study extensively by increasingly in-depth study；It with acoustics,
The tight phase in the subjects such as linguistics, psychology, signal transacting, artificial intelligence, pattern-recognition, information theory and computer field
Even；It shows huge application prospect in many fields, and many high performance speech recognition systems are also come out one after another；
Meanwhile, man-machine interaction is made by way of natural language, have far-reaching significance, be widely applied prospect and application field；It is first
First, the intelligent sound input based on mode identification technology, can bring revolutionary impact to office automation；Secondly, voice
Identification technology will greatly reduce the cumbersome and work of dullness in the extensive use in service industry field, save substantial amounts of manpower, carry
High workload efficiency；Again, speech recognition can also embody its powerful advantage on dangerous, severe working environment and battlefield；Cause
This, the research work of speech recognition for improving people's living standard, strengthen the various aspects such as national defense construction suffer from it is far-reaching
Meaning.
The content of the invention
Based on the technical problem that background technology is present, the present invention proposes a kind of automatic speech recognition based on dsp chip
System.
A kind of automatic speech recognition system based on dsp chip proposed by the present invention, including it is voice signal acquisition device, small
Wave filter, speech signal pre-processing module, speech recognition module, neural network module, Pattern Matching Module, language
Sound recognizes output module and dsp chip, the voice signal acquisition device, wavelet filter, speech signal pre-processing module, language
Message characteristic extracting module, Pattern Matching Module and speech recognition output module are sequentially connected, and speech recognition
Module and Pattern Matching Module are connected with neural network module, the voice signal acquisition device, wavelet filter, voice letter
Number pretreatment module, speech recognition module, neural network module, Pattern Matching Module and speech recognition output module
It is connected with dsp chip.
Preferably, the speech signal pre-processing module includes pre-emphasis unit, windowing unit and end-point detection unit, institute
State pre-emphasis unit, windowing unit and end-point detection unit to be sequentially connected, pre-emphasis unit is connected with wavelet filter, and end points
Detection unit is connected with speech recognition module, and pre-emphasis unit is preaccentuator.
Preferably, the neural network module is including training unit, modeling unit and infers unit, the training unit,
Modeling unit and deduction unit are sequentially connected, and training unit is connected with speech recognition module, and infers unit and mould
Formula matching module is connected.
Preferably, the wavelet filter is used to choose the useful information of voice signal, and suppresses irrelevant information to knowing
Not produced interference, speech signal pre-processing module is used to remove the voice signal of non-speech segment, speech recognition
Module is used to for pretreated voice signal to extract effective argument sequence for neural network module and Pattern Matching Module
Use.
In the present invention, the automatic speech recognition system that should be based on dsp chip can choose voice letter by wavelet filter
Number useful information, and suppress irrelevant information to the interference produced by identification, can be gone by speech signal pre-processing module
Except the voice signal of non-speech segment, can be to pretreated voice signal by time domain by speech recognition module
And frequency-domain analysis, extract effective argument sequence and used for neural network module and Pattern Matching Module, by neutral net
Module can summarize the rule of speech recognition, the voice signal of input can be carried out according to rule by Pattern Matching Module
Match somebody with somebody, reach the purpose of identification, the present invention is based on Speech processing, small echo and neural network theory and method, have studied language
The Dynamic Recognition of message number, speech recognition is applied to by small echo and neural network theory with method, and voice can be carried out automatically
Identification, simple structure is easy to use, low cost.
Brief description of the drawings
Fig. 1 is a kind of structural representation of automatic speech recognition system based on dsp chip proposed by the present invention.
Specific embodiment
The present invention is made with reference to specific embodiment further explain.
Embodiment
With reference to Fig. 1, the present embodiment proposes a kind of automatic speech recognition system based on dsp chip, including voice signal
Acquisition device, wavelet filter, speech signal pre-processing module, speech recognition module, neural network module, pattern
Matching module, speech recognition output module and dsp chip, voice signal acquisition device, wavelet filter, speech signal pre-processing
Module, speech recognition module, Pattern Matching Module and speech recognition output module are sequentially connected, and voice signal is special
Levy extraction module and Pattern Matching Module to be connected with neural network module, voice signal acquisition device, wavelet filter, voice
Signal pre-processing module, speech recognition module, neural network module, Pattern Matching Module and speech recognition output mould
Block is connected with dsp chip, and the automatic speech recognition system that should be based on dsp chip can choose voice letter by wavelet filter
Number useful information, and suppress irrelevant information to the interference produced by identification, can be gone by speech signal pre-processing module
Except the voice signal of non-speech segment, can be to pretreated voice signal by time domain by speech recognition module
And frequency-domain analysis, extract effective argument sequence and used for neural network module and Pattern Matching Module, by neutral net
Module can summarize the rule of speech recognition, the voice signal of input can be carried out according to rule by Pattern Matching Module
Match somebody with somebody, reach the purpose of identification, the present invention is based on Speech processing, small echo and neural network theory and method, have studied language
The Dynamic Recognition of message number, speech recognition is applied to by small echo and neural network theory with method, and voice can be carried out automatically
Identification, simple structure is easy to use, low cost.
In the present embodiment, speech signal pre-processing module includes pre-emphasis unit, windowing unit and end-point detection unit, in advance
Weighting unit, windowing unit and end-point detection unit are sequentially connected, and pre-emphasis unit is connected with wavelet filter, and end-point detection
Unit is connected with speech recognition module, and pre-emphasis unit is preaccentuator, neural network module include training unit,
Modeling unit and infer unit, training unit, modeling unit and infer that unit is sequentially connected, training unit and phonic signal character
Extraction module is connected, and infers that unit is connected with Pattern Matching Module, and wavelet filter is used to choose the useful letter of voice signal
Breath, and suppress irrelevant information to the interference produced by identification, speech signal pre-processing module is used to remove the language of non-speech segment
Message number, speech recognition module is used to for pretreated voice signal to extract effective argument sequence for nerve
Mixed-media network modules mixed-media and Pattern Matching Module are used, and the automatic speech recognition system that should be based on dsp chip can by wavelet filter
The useful information of voice signal is chosen, and suppresses irrelevant information to the interference produced by identification, by speech signal pre-processing
Module can remove the voice signal of non-speech segment, and pretreated voice can be believed by speech recognition module
Number by time and frequency domain analysis, extract effective argument sequence and used for neural network module and Pattern Matching Module, lead to
Crossing neural network module can summarize the rule of speech recognition, by Pattern Matching Module be capable of will be input into voice signal according to
Rule is matched, and reaches the purpose of identification, and the present invention is based on Speech processing, small echo and neural network theory and side
Method, have studied the Dynamic Recognition of voice signal, and small echo and neural network theory are applied into speech recognition with method, can be automatic
Voice is identified, simple structure is easy to use, low cost.
In the present embodiment, voice signal acquisition device obtains voice signal, is then transmit to wavelet filter, wavelet filtering
Device chooses the useful information of voice signal, and suppresses irrelevant information to the interference produced by identification, then passes voice signal
Speech signal pre-processing module is transported to, the effect of pre-emphasis unit is, by high boost, to be produced when lip is radiated to make up sound
Raw high frequency loss；, by digitized voice signal s (n) by a low-order digit system, this digital display circuit can be for it
Fixed, or slow self adaptation；Preaccentuator uses the first-order system of most widely used fixation, and its transmission function is such as
Under：
Here output s ' (n) of preemphasis is related by the input of following difierence equation to system：
The conventional window function of windowing unit has rectangular window, Hamming window and Hanning window etc., due to Hamming in actual application
The frequency characteristic of window is more suitable for the analysis of voice signal, so the system is weighted using Hamming window to signal, Hamming window
Function formula it is as follows：
Its frequency characteristic is：
End-point detection unit：Several seconds voice to collecting and recording must make end-point detection to distinguish sound section and unvoiced segments, can
The foundation for realizing end points judgement be voice of different nature various parameters in short-term have different probability density function and
Adjacent some frame voices should have consistent characteristics of speech sounds；Then speech signal pre-processing module believes pretreated voice
Number transmit to speech recognition module, speech recognition module is fallen by linear predictor coefficient and linear prediction
Spectral coefficient carries out feature extraction, and linear predictor coefficient is the linear prediction of voice, and its basic thought is：Each of voice signal takes
Sample value, can be represented with the weighted sum linear combination of its past several sampling value；The determination principle of each weight coefficient is
Make the mean-square value of predicated error minimum.
If be predicted using p sampling value of past, referred to as the linear prediction of p ranks；If with p sampling value { x of past
(n-1), x (n-2) ..., x (n-p) } weighting carry out prediction signal current sample value, then predicted value has x (n)：
Wherein, weight coefficient use-aplRepresent, referred to as predictive coefficient；Predicated error is：
Make predictive coefficient optimal, even if
ε=E [e2(n)]=min
Predictive coefficient can be solved by Durbin recursive algorithms, be comprised the following steps that：Iterative calculation is to be opened from p=0 from zeroth order
Begin；Zeroth order prediction does not give a forecast, and at this moment predicts that multinomial is
A0(z)=1
Predicated error is
e0(n)=x (n)
Predicated error power is
This is the primary condition of iterative calculation；Iterative step is as follows：
1. initialize
2. the parameter of known p rank fallout predictors, i.e., known A are assumedP(z) and εp；
3. the reflectance factor of p+1 rank fallout predictors is calculated：
4. the predictive coefficient of p+1 rank fallout predictors is calculated：
The prediction multinomial of corresponding p+1 ranks fallout predictor is：
Ap+1(z)=Ap(z)-γp+1z-(p+1)Ap(z-1)
5. p+1 rank predicated error power is calculated：
6. the is returned 2. to walk.
After calculating terminates, following three classes result is obtained：The predictive coefficient of each rank fallout predictor；The reflection system of each rank fallout predictor
Number；Each rank predicated error power.
Linear prediction residue error：Because voice signal has short-term stationarity, therefore characteristics of speech sounds also can use short-time spectrum
Represent, cepstrum is conventional one kind；Cepstrum is that inverse Fourier of the signal after Fourier transform gained power spectrum is taken the logarithm becomes
Change；Can be separated for recurrent pulse and sound channel by it, that is, obtain channel parameters；Cepstrum coefficient can directly be tried to achieve by the definition of cepstrum,
Also can be obtained by LPC coefficient recursion；Compared with directly cepstrum coefficient is calculated, the amount of calculation drug effect of LPCCEP, therefore the system is used
LPC cepstrum coefficients；There is a kind of very simple and effective recursive algorithm in the cepstrum based on lpc analysis：
In formula, CmIt is cepstrum coefficient, apIt is predictive coefficient, m is the exponent number (m=1-Q) of cepstrum coefficient, and p is predictive coefficient
Exponent number；The phonic signal character of extraction is transmitted to neural network module and carries out speech recognition regularity summarization, while the voice for extracting
Signal characteristic is transmitted to Pattern Matching Module, and the speech recognition rule that Pattern Matching Module is summarized according to neural network module is to defeated
The voice signal for entering carries out match cognization.
The above, the only present invention preferably specific embodiment, but protection scope of the present invention is not limited thereto,
Any one skilled in the art the invention discloses technical scope in, technology according to the present invention scheme and its
Inventive concept is subject to equivalent or change, should all be included within the scope of the present invention.
Claims (4)
1. a kind of automatic speech recognition system based on dsp chip, including voice signal acquisition device, wavelet filter, voice
Signal pre-processing module, speech recognition module, neural network module, Pattern Matching Module, speech recognition output mould
Block and dsp chip, it is characterised in that the voice signal acquisition device, wavelet filter, speech signal pre-processing module, language
Message characteristic extracting module, Pattern Matching Module and speech recognition output module are sequentially connected, and speech recognition
Module and Pattern Matching Module are connected with neural network module, the voice signal acquisition device, wavelet filter, voice letter
Number pretreatment module, speech recognition module, neural network module, Pattern Matching Module and speech recognition output module
It is connected with dsp chip.
2. a kind of automatic speech recognition system based on dsp chip according to claim 1, it is characterised in that institute's predicate
Sound signal pre-processing module includes pre-emphasis unit, windowing unit and end-point detection unit, the pre-emphasis unit, windowing unit
It is sequentially connected with end-point detection unit, pre-emphasis unit is connected with wavelet filter, and end-point detection unit is special with voice signal
Extraction module connection is levied, pre-emphasis unit is preaccentuator.
3. a kind of automatic speech recognition system based on dsp chip according to claim 1, it is characterised in that the god
Include training unit, modeling unit through mixed-media network modules mixed-media and infer unit, the training unit, modeling unit and deduction unit are successively
Connection, training unit is connected with speech recognition module, and infers that unit is connected with Pattern Matching Module.
4. a kind of automatic speech recognition system based on dsp chip according to claim 1, it is characterised in that described small
Wave filter is used to choose the useful information of voice signal, and suppresses irrelevant information to the interference produced by identification, voice letter
Number pretreatment module is used to remove the voice signal of non-speech segment, and speech recognition module is used for pretreated language
Message number extracts effective argument sequence and is used for neural network module and Pattern Matching Module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611064684.7A CN106782550A (en) | 2016-11-28 | 2016-11-28 | A kind of automatic speech recognition system based on dsp chip |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611064684.7A CN106782550A (en) | 2016-11-28 | 2016-11-28 | A kind of automatic speech recognition system based on dsp chip |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106782550A true CN106782550A (en) | 2017-05-31 |
Family
ID=58902279
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611064684.7A Pending CN106782550A (en) | 2016-11-28 | 2016-11-28 | A kind of automatic speech recognition system based on dsp chip |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106782550A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108601180A (en) * | 2018-06-04 | 2018-09-28 | 长江大学 | Electric light audio-switch, control system and method based on sound groove recognition technology in e |
| CN110047480A (en) * | 2019-04-22 | 2019-07-23 | 哈尔滨理工大学 | Added Management robot head device and control for the inquiry of department, community hospital |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102789779A (en) * | 2012-07-12 | 2012-11-21 | 广东外语外贸大学 | Speech recognition system and recognition method thereof |
| CN103065629A (en) * | 2012-11-20 | 2013-04-24 | 广东工业大学 | Speech recognition system of humanoid robot |
| CN103236260A (en) * | 2013-03-29 | 2013-08-07 | 京东方科技集团股份有限公司 | Voice recognition system |
-
2016
- 2016-11-28 CN CN201611064684.7A patent/CN106782550A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102789779A (en) * | 2012-07-12 | 2012-11-21 | 广东外语外贸大学 | Speech recognition system and recognition method thereof |
| CN103065629A (en) * | 2012-11-20 | 2013-04-24 | 广东工业大学 | Speech recognition system of humanoid robot |
| CN103236260A (en) * | 2013-03-29 | 2013-08-07 | 京东方科技集团股份有限公司 | Voice recognition system |
Non-Patent Citations (6)
| Title |
|---|
| 关胜平: ""基于TMS320VC5509A的语音识别与控制***"", 《电子技术应用》 * |
| 曹斌芳: ""一种采用小波变换的实时语音识别***设计"", 《INTERNATIONAL CONFERENCE ON POWER ELECTRONICS & INTELLIGENT TRANSPORTATION SYSTEM》 * |
| 李子琳: ""语音信号识别技术及应用研究"", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 李记昌: ""基于DSP的语音处理及识别算法研究"", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 田丽: ""小波预处理在语音识别***中的应用"", 《科技创新导报》 * |
| 闫文娟: ""基于TMS320C5409的语音识别***"", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108601180A (en) * | 2018-06-04 | 2018-09-28 | 长江大学 | Electric light audio-switch, control system and method based on sound groove recognition technology in e |
| CN110047480A (en) * | 2019-04-22 | 2019-07-23 | 哈尔滨理工大学 | Added Management robot head device and control for the inquiry of department, community hospital |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP2695160B1 (en) | Speech syllable/vowel/phone boundary detection using auditory attention cues | |
| CN104008751A (en) | Speaker recognition method based on BP neural network | |
| CN103117059A (en) | Voice signal characteristics extracting method based on tensor decomposition | |
| CN109192200B (en) | Speech recognition method | |
| CN113012720B (en) | Depression detection method by multi-voice feature fusion under spectral subtraction noise reduction | |
| Ganapathy et al. | Feature extraction using 2-d autoregressive models for speaker recognition. | |
| Chaudhary et al. | Gender identification based on voice signal characteristics | |
| CN113077806B (en) | Audio processing method and device, model training method and device, medium and equipment | |
| CN112786059A (en) | Voiceprint feature extraction method and device based on artificial intelligence | |
| Mistry et al. | Overview: Speech recognition technology, mel-frequency cepstral coefficients (mfcc), artificial neural network (ann) | |
| CN110136726A (en) | A kind of estimation method, device, system and the storage medium of voice gender | |
| CN115510909A (en) | Unsupervised algorithm for DBSCAN to perform abnormal sound features | |
| Labied et al. | An overview of automatic speech recognition preprocessing techniques | |
| CN112183582A (en) | Multi-feature fusion underwater target identification method | |
| Labied et al. | Automatic speech recognition features extraction techniques: A multi-criteria comparison | |
| Mu et al. | Voice activity detection optimized by adaptive attention span transformer | |
| CN106782550A (en) | A kind of automatic speech recognition system based on dsp chip | |
| CN112634880A (en) | Speaker identification method, device, equipment, storage medium and program product | |
| Sundar et al. | A mixture model approach for formant tracking and the robustness of student's-t distribution | |
| CN114913859B (en) | Voiceprint recognition method, voiceprint recognition device, electronic equipment and storage medium | |
| Liu et al. | Pitch-synchronous linear prediction analysis of high-pitched speech using weighted short-time energy function | |
| Chazan et al. | Efficient periodicity extraction based on sine-wave representation and its application to pitch determination of speech signals. | |
| KR20180101057A (en) | Method and apparatus for voice activity detection robust to noise | |
| CN114550696A (en) | Method and system for realizing emotion judgment through voice recognition | |
| CN113742515A (en) | Audio classification method, system, 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170531 |
|
| RJ01 | Rejection of invention patent application after publication |