CN110033757A - A kind of voice recognizer - Google Patents
A kind of voice recognizer Download PDFInfo
- Publication number
- CN110033757A CN110033757A CN201910272975.2A CN201910272975A CN110033757A CN 110033757 A CN110033757 A CN 110033757A CN 201910272975 A CN201910272975 A CN 201910272975A CN 110033757 A CN110033757 A CN 110033757A
- Authority
- CN
- China
- Prior art keywords
- speaker
- sound
- data
- algorithm
- characteristic
- 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
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; 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 OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/16—Speech classification or search using artificial neural networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
Abstract
The present invention provides a kind of voice recognizer, comprising the following steps: S1: to the self-adaptive processing of speaker's wave volume size, doing overall normalized to identical maximum value after the wave volume size of speaker is carried out identification model training;S2: to the self-adaptive processing in the mute area of speaker's sound, the volume numerical values recited of current speaker is judged by mean filter, then falls mute region by threshold filtering;S3: noise reduction is filtered to background sound, and consistency treatment is done to speaker's voice data;S4: extracting the sound characteristic of speaker, by trained neural network algorithm model, extracts the high dimensional feature vector of speaker's sound;S5: being compared identification with vocal print library to the sound characteristic of speaker, and the high dimensional feature of neural network algorithm model extraction is compared using COS distance, obtains the similarity of speaker characteristic.The present invention directly recognizes the sound characteristic of speaker, and noise is small, and arithmetic accuracy is high.
Description
Technical field
The invention belongs to voice recognition technology fields, and in particular to a kind of voice recognizer.
Background technique
Speech recognition technology is to be converted into sound, byte or the phrase that human hair goes out by the identification and understanding process of machine
Corresponding text or symbol, or provide a kind of information technology of response.With the rapid development of information technology, speech recognition skill
Art has been widely used in daily life.For example, can be passed through when using terminal equipment using speech recognition technology
The mode of input voice easily inputs information in terminal device.
Speech recognition technology be substantially one mode identification process, the ginseng of the mode of unknown voice and known voice
It examines mode to be compared one by one, the reference model of best match is exported as recognition result.Existing speech recognition technology is adopted
There are many recognition methods, such as model matching method, probabilistic model method etc..What industry generallyd use at present is probabilistic model method
Speech recognition technology.Probabilistic model method speech recognition technology is to be carried out by cloud to the voice that a large amount of different user inputs
Acoustics training, and a general acoustic model is obtained, it will be to be identified according to the general acoustic model and speech model
Voice signal is decoded as text output.This recognition methods can be to the language of most people for unspecified person
Sound identified, still, since it is general acoustic model, when user pronunciation is not up to standard, or has accent,
This general acoustic model just can not accurately carry out matching primitives, and substantive defect is can only to identify saying for speaker
Content is talked about, directly the sound characteristic of speaker directly can not be recognized.
Summary of the invention
The object of the present invention is to provide a kind of voice recognizers, are directly recognized, are made an uproar to the sound characteristic of speaker
Sound is small, and arithmetic accuracy is high.
The present invention provides the following technical solutions:
A kind of voice recognizer, comprising the following steps:
S1: to the self-adaptive processing of speaker's wave volume size, the wave volume size of speaker is subjected to identification mould
Overall normalized is done after type training to identical maximum value;
S2: to the self-adaptive processing in the mute area of speaker's sound, the volume of current speaker is judged by mean filter
Numerical values recited, then fall mute region by threshold filtering;
S3: noise reduction is filtered to background sound, and consistency treatment is done to speaker's voice data;
S4: extracting the sound characteristic of speaker, by trained neural network algorithm model, extracts speaker's sound
High dimensional feature vector;
S5: being compared identification with vocal print library to the sound characteristic of speaker, compares neural network using COS distance and calculates
The high dimensional feature of method model extraction obtains the similarity of speaker characteristic.
Preferably, the volume of input sound characteristic is used in the S1 to the self-adaptive processing of speaker's wave volume size
Size adaptation algorithm, it is described input sound characteristic volume adaptive algorithm the following steps are included:
S11: the maximum value of current sound data is found out;
S12: on the basis of volume maximum value, the coefficient value relative to a constant is found out;
S13: current sound data multiply the data that the coefficient obtains adaptive volume.
Preferably, the mute area of input sound characteristic is used in the S2 to the self-adaptive processing in the mute area of speaker's sound
From pruning algorithm, it is described input sound characteristic mute area from pruning algorithm the following steps are included:
S21: the mean value of the absolute value of 1600 sample magnitude about 0.1s is sought;
Whether S22: being mute area by threshold decision current data;
S23: truncation is carried out to the data in then mute area.
Preferably, include using noise reduction algorithm, the filtering noise reduction algorithm is filtered to background sound filtering noise reduction in the S3
Following steps:
S31: the extraction of background noise data is carried out using wavelet algorithm;
S32: being divided into multiband for band data of making an uproar, each frequency band and noise data seek matrix inner products, utilize orthogonal matrix
Product be zero principle, filter out the noise sample currently extracted.
Preferably, doing consistency treatment to speaker's voice data in the S3 is to the FBANK feature of speaker and one
Rank increment and second-order increment data operate after subtracting mean value divided by the standard deviation of data, obtain consistency voice data.
Preferably, the sound characteristic of speaker is extracted in the S4 by trained neural network algorithm model, is used
Multilayer convolution algorithm and full-join algorithm extract 512 high dimensional feature vector classification of sound of speaker.
The beneficial effects of the present invention are: this recognizer to pre-input sound characteristic processing so that input feature vector it is more perfect,
Noise is smaller, arithmetic accuracy is higher;Classification is extracted using the high dimensional feature that depth nerve convolutional network algorithm carries out sound, to saying
Words voice sound feature extraction directly directly recognizes the sound characteristic of speaker, avoid passing through the speech content of speaker into
The defect of row identification.
Detailed description of the invention
Attached drawing is used to provide further understanding of the present invention, and constitutes part of specification, with reality of the invention
It applies example to be used to explain the present invention together, not be construed as limiting the invention.In the accompanying drawings:
Fig. 1 is recognizer flow diagram of the present invention;
Fig. 2 is that deep neural network structural diagrams are intended to.
Specific embodiment
As shown in Figure 1, a kind of voice recognizer, comprising the following steps:
S1: to the self-adaptive processing of speaker's wave volume size, the wave volume size of speaker is subjected to identification mould
Overall normalized is done after type training to identical maximum value;
S2: to the self-adaptive processing in the mute area of speaker's sound, the volume of current speaker is judged by mean filter
Numerical values recited, then fall mute region by threshold filtering;
S3: noise reduction is filtered to background sound, and consistency treatment is done to speaker's voice data;
S4: extracting the sound characteristic of speaker, by trained neural network algorithm model, extracts speaker's sound
High dimensional feature vector;
S5: being compared identification with vocal print library to the sound characteristic of speaker, compares neural network using COS distance and calculates
The high dimensional feature of method model extraction obtains the similarity of speaker characteristic.
Specifically, to the self-adaptive processing of speaker's wave volume size using the volume of input sound characteristic in S1
Adaptive algorithm, input the volume adaptive algorithm of sound characteristic the following steps are included:
S11: the maximum value of current sound data is found out;
S12: on the basis of volume maximum value, the coefficient value relative to a constant is found out;
S13: current sound data multiply the data that coefficient obtains adaptive volume.
Specifically, being cut certainly to the self-adaptive processing in the mute area of speaker's sound using the mute area of input sound characteristic in S2
Disconnected algorithm, input the mute area of sound characteristic from pruning algorithm the following steps are included:
S21: the mean value of the absolute value of 1600 sample magnitude about 0.1s is sought;
Whether S22: being mute area by threshold decision current data;
S23: truncation is carried out to the data in then mute area.
Specifically, in S3 to background sound filtering noise reduction using filtering noise reduction algorithm, filtering noise reduction algorithm the following steps are included:
S31: the extraction of background noise data is carried out using wavelet algorithm;
S32: being divided into multiband for band data of making an uproar, each frequency band and noise data seek matrix inner products, utilize orthogonal matrix
Product be zero principle, filter out the noise sample currently extracted.
Specifically, doing consistency treatment to speaker's voice data in S3 is to increase to the FBANK feature and single order of speaker
Amount and second-order increment data operate after subtracting mean value divided by the standard deviation of data, obtain consistency voice data.
As depicted in figs. 1 and 2, the sound characteristic that speaker is extracted in S4 passes through trained neural network algorithm model,
512 high dimensional feature vector classification of sound of speaker are extracted using multilayer convolution algorithm and full-join algorithm.
As depicted in figs. 1 and 2, a kind of voice recognizer, specifically, convolution sum full-join algorithm: convolutional neural networks
Basic structure include two layers, one is characterized extract layer, and the input of each neuron is connected with the local acceptance region of preceding layer,
And extract the feature of the part.After the local feature is extracted, its positional relationship between other feature also determines therewith
Get off;The second is Feature Mapping layer, each computation layer of network is made of multiple Feature Mappings, and each Feature Mapping is one flat
Face, the weight of all neurons is equal in plane.Feature Mapping structure is using the small sigmoid function of influence function core as volume
The activation primitive of product network, so that Feature Mapping has shift invariant.Further, since the neuron on a mapping face is shared
Weight, thus reduce the number of network freedom parameter.Each of convolutional neural networks convolutional layer all followed by use
The computation layer of local average and second extraction is sought, this distinctive structure of feature extraction twice reduces feature resolution.
Convolutional neural networks are mainly used to the X-Y scheme of identification displacement, scaling and other forms distortion invariance.Due to
The feature detection layer of convolutional neural networks is learnt by training data, so avoiding when using convolutional neural networks
The feature extraction of display, and implicitly learnt from training data;Furthermore due to the neuron on same Feature Mapping face
Weight is identical, so network can be with collateral learning, this is also convolutional network is connected with each other the one big excellent of network relative to neuron
Gesture.Convolutional neural networks have in terms of speech recognition and image procossing unique excellent with the special construction that its local weight is shared
More property, layout share the complexity for reducing network closer to actual biological neural network, weight, and especially multidimensional is defeated
The image of incoming vector can directly input the complexity that network this feature avoids data reconstruction in feature extraction and assorting process
Degree, is being eventually adding full articulamentum, in this way may be used in full articulamentum using the output of the 4th layer of convolution as the input of full articulamentum
To learn to part and global feature.The height that the present invention mainly uses multilayer convolution algorithm that full-join algorithm is added to carry out sound
Dimensional feature extracts classification, to directly directly be recognized to the sound characteristic of speaker, it is interior to avoid passing through speaking for speaker
Hold the defect identified, arithmetic accuracy is higher.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, although referring to aforementioned reality
Applying example, invention is explained in detail, for those skilled in the art, still can be to aforementioned each implementation
Technical solution documented by example is modified or equivalent replacement of some of the technical features.It is all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (6)
1. a kind of voice recognizer, which comprises the following steps:
S1: to the self-adaptive processing of speaker's wave volume size, the wave volume size of speaker is subjected to identification model instruction
Overall normalized is done after white silk to identical maximum value;
S2: to the self-adaptive processing in the mute area of speaker's sound, the volume numerical value of current speaker is judged by mean filter
Size, then fall mute region by threshold filtering;
S3: noise reduction is filtered to background sound, and consistency treatment is done to speaker's voice data;
S4: extracting the sound characteristic of speaker, by trained neural network algorithm model, extracts the higher-dimension of speaker's sound
Feature vector;
S5: being compared identification with vocal print library to the sound characteristic of speaker, compares neural network algorithm mould using COS distance
The high dimensional feature that type extracts, obtains the similarity of speaker characteristic.
2. a kind of voice recognizer according to claim 1, which is characterized in that speaker's wave volume in the S1
The self-adaptive processing of size is given great volume using the volume adaptive algorithm for inputting sound characteristic, the input sound characteristic
Small adaptive algorithm the following steps are included:
S11: the maximum value of current sound data is found out;
S12: on the basis of volume maximum value, the coefficient value relative to a constant is found out;
S13: current sound data multiply the data that the coefficient obtains adaptive volume.
3. a kind of voice recognizer according to claim 1, which is characterized in that mute to speaker's sound in the S2
Using the mute area of input sound characteristic from pruning algorithm, the mute area of the input sound characteristic is cut the self-adaptive processing in area certainly
Disconnected algorithm the following steps are included:
S21: the mean value of the absolute value of 1600 sample magnitude about 0.1s is sought;
Whether S22: being mute area by threshold decision current data;
S23: truncation is carried out to the data in then mute area.
4. a kind of voice recognizer according to claim 1, which is characterized in that filter noise reduction to background sound in the S3
Using filtering noise reduction algorithm, the filtering noise reduction algorithm the following steps are included:
S31: the extraction of background noise data is carried out using wavelet algorithm;
S32: being divided into multiband for band data of making an uproar, each frequency band and noise data seek matrix inner products, utilize the product of orthogonal matrix
The principle for being zero filters out the noise sample currently extracted.
5. a kind of voice recognizer according to claim 1, which is characterized in that speaker's voice data in the S3
Doing consistency treatment is after subtract to the FBANK feature and first increment and second-order increment data of speaker mean value divided by number
According to standard deviation operation, obtain consistency voice data.
6. a kind of voice recognizer according to claim 1, which is characterized in that extract the sound of speaker in the S4
Feature extracts speaker's using multilayer convolution algorithm and full-join algorithm by trained neural network algorithm model
512 high dimensional feature vector classification of sound.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910272975.2A CN110033757A (en) | 2019-04-04 | 2019-04-04 | A kind of voice recognizer |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910272975.2A CN110033757A (en) | 2019-04-04 | 2019-04-04 | A kind of voice recognizer |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110033757A true CN110033757A (en) | 2019-07-19 |
Family
ID=67237513
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910272975.2A Pending CN110033757A (en) | 2019-04-04 | 2019-04-04 | A kind of voice recognizer |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110033757A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110534127A (en) * | 2019-09-24 | 2019-12-03 | 华南理工大学 | Applied to the microphone array voice enhancement method and device in indoor environment |
CN112509555A (en) * | 2020-11-25 | 2021-03-16 | 平安科技(深圳)有限公司 | Dialect voice recognition method, dialect voice recognition device, dialect voice recognition medium and electronic equipment |
CN112951266A (en) * | 2021-02-05 | 2021-06-11 | 杭州网易云音乐科技有限公司 | Tooth sound adjusting method, tooth sound adjusting device, electronic equipment and computer readable storage medium |
CN115967894A (en) * | 2022-12-15 | 2023-04-14 | 广州迅控电子科技有限公司 | Microphone sound processing method, system, terminal equipment and storage medium |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102136272A (en) * | 2010-01-26 | 2011-07-27 | 雅马哈株式会社 | Masker sound generation apparatus |
CN103778917A (en) * | 2014-01-10 | 2014-05-07 | 厦门快商通信息技术有限公司 | System and method for detecting identity impersonation in telephone satisfaction survey |
CN103971690A (en) * | 2013-01-28 | 2014-08-06 | 腾讯科技(深圳)有限公司 | Voiceprint recognition method and device |
CN104269178A (en) * | 2014-08-08 | 2015-01-07 | 华迪计算机集团有限公司 | Method and device for conducting self-adaption spectrum reduction and wavelet packet noise elimination processing on voice signals |
CN104732978A (en) * | 2015-03-12 | 2015-06-24 | 上海交通大学 | Text-dependent speaker recognition method based on joint deep learning |
CN105118511A (en) * | 2015-07-31 | 2015-12-02 | 国网电力科学研究院武汉南瑞有限责任公司 | Thunder identification method |
CN106935248A (en) * | 2017-02-14 | 2017-07-07 | 广州孩教圈信息科技股份有限公司 | A kind of voice similarity detection method and device |
CN107492382A (en) * | 2016-06-13 | 2017-12-19 | 阿里巴巴集团控股有限公司 | Voiceprint extracting method and device based on neutral net |
CN107886943A (en) * | 2017-11-21 | 2018-04-06 | 广州势必可赢网络科技有限公司 | A kind of method for recognizing sound-groove and device |
CN108648759A (en) * | 2018-05-14 | 2018-10-12 | 华南理工大学 | A kind of method for recognizing sound-groove that text is unrelated |
CN108711436A (en) * | 2018-05-17 | 2018-10-26 | 哈尔滨工业大学 | Speaker verification's system Replay Attack detection method based on high frequency and bottleneck characteristic |
CN108766445A (en) * | 2018-05-30 | 2018-11-06 | 苏州思必驰信息科技有限公司 | Method for recognizing sound-groove and system |
CN108986825A (en) * | 2018-07-02 | 2018-12-11 | 北京百度网讯科技有限公司 | Context acquisition methods and equipment based on interactive voice |
CN109243488A (en) * | 2018-10-30 | 2019-01-18 | 腾讯音乐娱乐科技(深圳)有限公司 | Audio-frequency detection, device and storage medium |
-
2019
- 2019-04-04 CN CN201910272975.2A patent/CN110033757A/en active Pending
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102136272A (en) * | 2010-01-26 | 2011-07-27 | 雅马哈株式会社 | Masker sound generation apparatus |
CN103971690A (en) * | 2013-01-28 | 2014-08-06 | 腾讯科技(深圳)有限公司 | Voiceprint recognition method and device |
CN103778917A (en) * | 2014-01-10 | 2014-05-07 | 厦门快商通信息技术有限公司 | System and method for detecting identity impersonation in telephone satisfaction survey |
CN104269178A (en) * | 2014-08-08 | 2015-01-07 | 华迪计算机集团有限公司 | Method and device for conducting self-adaption spectrum reduction and wavelet packet noise elimination processing on voice signals |
CN104732978A (en) * | 2015-03-12 | 2015-06-24 | 上海交通大学 | Text-dependent speaker recognition method based on joint deep learning |
CN105118511A (en) * | 2015-07-31 | 2015-12-02 | 国网电力科学研究院武汉南瑞有限责任公司 | Thunder identification method |
CN107492382A (en) * | 2016-06-13 | 2017-12-19 | 阿里巴巴集团控股有限公司 | Voiceprint extracting method and device based on neutral net |
CN106935248A (en) * | 2017-02-14 | 2017-07-07 | 广州孩教圈信息科技股份有限公司 | A kind of voice similarity detection method and device |
CN107886943A (en) * | 2017-11-21 | 2018-04-06 | 广州势必可赢网络科技有限公司 | A kind of method for recognizing sound-groove and device |
CN108648759A (en) * | 2018-05-14 | 2018-10-12 | 华南理工大学 | A kind of method for recognizing sound-groove that text is unrelated |
CN108711436A (en) * | 2018-05-17 | 2018-10-26 | 哈尔滨工业大学 | Speaker verification's system Replay Attack detection method based on high frequency and bottleneck characteristic |
CN108766445A (en) * | 2018-05-30 | 2018-11-06 | 苏州思必驰信息科技有限公司 | Method for recognizing sound-groove and system |
CN108986825A (en) * | 2018-07-02 | 2018-12-11 | 北京百度网讯科技有限公司 | Context acquisition methods and equipment based on interactive voice |
CN109243488A (en) * | 2018-10-30 | 2019-01-18 | 腾讯音乐娱乐科技(深圳)有限公司 | Audio-frequency detection, device and storage medium |
Non-Patent Citations (1)
Title |
---|
杨再礼等: "《广播技术基础概论》", 30 November 2014 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110534127A (en) * | 2019-09-24 | 2019-12-03 | 华南理工大学 | Applied to the microphone array voice enhancement method and device in indoor environment |
CN112509555A (en) * | 2020-11-25 | 2021-03-16 | 平安科技(深圳)有限公司 | Dialect voice recognition method, dialect voice recognition device, dialect voice recognition medium and electronic equipment |
CN112509555B (en) * | 2020-11-25 | 2023-05-23 | 平安科技(深圳)有限公司 | Dialect voice recognition method, device, medium and electronic equipment |
CN112951266A (en) * | 2021-02-05 | 2021-06-11 | 杭州网易云音乐科技有限公司 | Tooth sound adjusting method, tooth sound adjusting device, electronic equipment and computer readable storage medium |
CN112951266B (en) * | 2021-02-05 | 2024-02-06 | 杭州网易云音乐科技有限公司 | Tooth sound adjusting method, tooth sound adjusting device, electronic equipment and computer readable storage medium |
CN115967894A (en) * | 2022-12-15 | 2023-04-14 | 广州迅控电子科技有限公司 | Microphone sound processing method, system, terminal equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110033757A (en) | A kind of voice recognizer | |
Bai et al. | Speaker recognition based on deep learning: An overview | |
CN105047194B (en) | A kind of self study sound spectrograph feature extracting method for speech emotion recognition | |
CN112466326B (en) | Voice emotion feature extraction method based on transducer model encoder | |
CN106504768B (en) | Phone testing audio frequency classification method and device based on artificial intelligence | |
Demircan et al. | Feature extraction from speech data for emotion recognition | |
Alsobhani et al. | Speech recognition using convolution deep neural networks | |
Kumbhar et al. | Speech emotion recognition using MFCC features and LSTM network | |
La Mura et al. | Human-machine interaction personalization: a review on gender and emotion recognition through speech analysis | |
CN111128240B (en) | Voice emotion recognition method based on anti-semantic-erasure | |
US20230298616A1 (en) | System and Method For Identifying Sentiment (Emotions) In A Speech Audio Input with Haptic Output | |
US20210193159A1 (en) | Training a voice morphing apparatus | |
Shahzadi et al. | Recognition of emotion in speech using spectral patterns | |
Zhiyan et al. | Speech emotion recognition based on deep learning and kernel nonlinear PSVM | |
Singh et al. | Speech emotion recognition system using gender dependent convolution neural network | |
Ribeiro et al. | Binary neural networks for classification of voice commands from throat microphone | |
CN114463688A (en) | Cross-modal context coding dialogue emotion recognition method and system | |
Sailor et al. | Unsupervised Deep Auditory Model Using Stack of Convolutional RBMs for Speech Recognition. | |
JP2015175859A (en) | Pattern recognition device, pattern recognition method, and pattern recognition program | |
CN116230020A (en) | Speech emotion recognition and classification method | |
CN115145402A (en) | Intelligent toy system with network interaction function and control method | |
Mohamed et al. | Speech recognition system based on discrete wave atoms transform partial noisy environment | |
CN111785262B (en) | Speaker age and gender classification method based on residual error network and fusion characteristics | |
Kovács et al. | The joint optimization of spectro-temporal features and neural net classifiers | |
CN114881668A (en) | Multi-mode-based deception detection method |
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 | ||
TA01 | Transfer of patent application right | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20191204 Address after: 210000 Room 201, 33 Tianmu Road, Gulou District, Nanjing City, Jiangsu Province Applicant after: Li Dongming Address before: Jiangning District of Nanjing City, Jiangsu province 210000 streets moling Road No. 12 mo Zhou Applicant before: Xingzhi Technology Co. Ltd. |
|
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190719 |