CN103903624B - Periodical pitch detection method under a kind of gauss heat source model environment - Google Patents
Periodical pitch detection method under a kind of gauss heat source model environment Download PDFInfo
- Publication number
- CN103903624B CN103903624B CN201410124483.6A CN201410124483A CN103903624B CN 103903624 B CN103903624 B CN 103903624B CN 201410124483 A CN201410124483 A CN 201410124483A CN 103903624 B CN103903624 B CN 103903624B
- Authority
- CN
- China
- Prior art keywords
- voice
- heat source
- frame
- storehouse
- source model
- 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.)
- Expired - Fee Related
Links
Abstract
The present invention provides the Periodical pitch detection method under a kind of gauss heat source model environment. It is characterized in that utilizing the voice storehouse (voice storehouse B) containing gauss heat source model to construct 4 rank semi-invariant diagonal slices vectors, the voice storehouse under quiet environment (voice storehouse A) is utilized to extract fundamental tone cycle parameter, 4 rank semi-invariant diagonal slices vectors are trained with the input and output of fundamental tone cycle parameter as generalized regression nerve networks learning sample, again using the 4 vectorial inputs as the GRNNs trained of rank semi-invariant diagonal slices of input voice frame, namely the output obtaining neural network GRNNs input the fundamental tone cycle of voice frame.
Description
Technical field
The present invention relates to Periodical pitch detection method, particularly the Periodical pitch detection method under a kind of gauss heat source model environment.
Background technology
The fundamental tone cycle, in compress speech, there was purposes widely in the speech processes fields such as voice analysis synthesis and speech recognition as the basic parameter of voice. Accurately and reliably estimating and to extract the fundamental tone cycle most important to Speech processing, it directly can affect the quality of synthetic speech, reduces the verity of voice and naturalness in phonetic recognization rate and speech coding and decoding system. Pitch determination mainly contains auto-relativity function method, average magnitude difference function method and the method for falling spectrum etc., but these methods are difficult to better effects under low signal-to-noise ratio environment, many innovatory algorithm are had in recent years for the pitch determination in noise environment, mostly make use of autocorrelative function, and autocorrelative function can only suppress white Gaussian noise, it is invalid to gauss heat source model. Given this, the present invention provides a kind of special in the Periodical pitch detection method under gauss heat source model.
Summary of the invention
Have obvious deficiency for the pitch determination that carries out of prior art under gauss heat source model, the present invention provide a kind of utilize the fourth-order cumulant diagonal slice vector of voice frame to carry out gauss heat source model under Periodical pitch detection method.
The method comprises the following steps:
(1) being reset under gauss heat source model environment the voice storehouse B made under gauss heat source model environment by the voice storehouse A under quiet environment, voice storehouse A is the some set of voice digital sample, voice storehouse B is the some set of voice digital sample, wherein L is total number of sample points;
(2) respectively to the speech signal sampling point temporally order framing in voice storehouse 1 and voice storehouse 2, paired voice frame is obtained
;
WhereinFor the voice frame of voice storehouse A,
WhereinFor the voice frame of voice storehouse B,
Wherein N is voice frame length, and i is voice frame ordinal number;
(3) voice frame is calculated4 rank semi-invariant diagonal slicesWhereinFor the voice frame sampling periodIntegral multiple, and construct 4 rank semi-invariant diagonal slices vectors of the i-th frame voice frame, and do stdn and can obtain
;
(4) voice frame is estimatedFundamental tone cycle parameter, and be designated as;
(5) will,Input and output as generalized regression nerve networks GRNNs learning sample are trained, n be input and output sample to sum, its core widthForSquare root of the variance;
(6) to the temporally order framing of input speech signal sampling point, and stdn 4 rank semi-invariant diagonal slices corresponding with it is calculated
;
(7) willIt is input in the GRNNs trained, can fundamental tone cycle of this voice frame��
The technique scheme of the present invention, compared with prior art, has the following advantages:
A, utilize voice frame fourth-order cumulant diagonal slice vector suppress gauss heat source model can have good effect;
The generalized regression nerve networks that B, utilization train estimates the fundamental tone cycle, it is possible to possess the performance advantage of precision and speed simultaneously;
Accompanying drawing explanation
Fig. 1 is the method flow diagram of the present invention
Embodiment
The present invention propose gauss heat source model environment under Periodical pitch detection method by reference to the accompanying drawings and embodiment be described as follows further:
The method flow of the present invention as shown in Figure 1, comprises the following steps:
(1) reset under gauss heat source model environment the voice storehouse B made under gauss heat source model environment by the voice storehouse A under quiet environment;
(2) respectively to the speech signal sampling point temporally order framing of voice storehouse A and voice storehouse B, paired voice frame is obtained
;
WhereinFor the voice frame of voice storehouse A,The voice frame of voice storehouse B, i is voice frame ordinal number;
(3) voice frame is calculated4 rank semi-invariant diagonal slicesWhereinFor the voice frame sampling periodIntegral multiple, and construct 4 rank semi-invariant diagonal slices vectors of the i-th frame voice frame, and do stdn and can obtain
;
(4) voice frame is estimatedFundamental tone cycle parameter, and be designated as;
(5) will,Input and output as generalized regression nerve networks GRNNs learning sample are trained, n be input and output sample to sum, its core widthForSquare root of the variance;
(6) to the temporally order framing of input speech signal sampling point, and stdn 4 rank semi-invariant diagonal slices corresponding with it is calculated;
(7) willIt is input in the GRNNs trained, can fundamental tone cycle of this voice frame��
The specific embodiment of each step of aforesaid method of the present invention is described in detail as follows:
The embodiment of the voice storehouse A in aforesaid method step (1) records the voice of China 30, main province ,city and area male sex and 30 women, and length 20 minutes during everyone voice, time total, length is 20 hours. The embodiment of voice storehouse B is superposition gauss heat source model on the basis of voice storehouse A.
The embodiment of voice storehouse A and voice storehouse B signal sampling point temporally order framing is by 8KHz frequency sampling by aforesaid method step (2), removes the voice sampling point of Hz noise through high pass. Every 25ms is also exactly that 200 voice sampling points form a frame.
In aforesaid method step (3), the embodiment of 4 rank semi-invariant diagonal slices vectors of stdn is 10 rank vectors
��
The embodiment of aforesaid method step (4) is: the fundamental tone cycle parameter asking for present frame by the method described by linear prediction (MELP) the speech coding algorithm standard of United States Government's 2400b/s mixed excitation.
The embodiment of aforesaid method step (5) is: by 4 rank semi-invariant diagonal slices vectors of stdnWith the fundamental tone cycleThe input and output of generalized regression nerve networks learning sample are trained, the core width of neural networkForSquare root of the variance.
Temporally sequentially the embodiment of framing is consistent with the embodiment of method steps (2) to input speech signal sampling point for aforesaid method step (6).
Aforesaid method step (7) embodiment is: by the stdn 4 rank semi-invariant diagonal slices of input speech signal sampling point voice frameAs the input of the generalized regression nerve networks trained in method steps (5), the fundamental tone cycle of this voice frame can be obtained��
Claims (2)
1. the Periodical pitch detection method under a gauss heat source model environment, it is characterised in that the method comprises the following steps:
(1) reset under gauss heat source model environment the voice storehouse B made under gauss heat source model environment by the voice storehouse A under quiet environment;
(2) respectively to the speech signal sampling point temporally order framing of voice storehouse A and voice storehouse B, paired voice frame is obtained
;
WhereinFor the voice frame of voice storehouse A,The voice frame of voice storehouse B, i is voice frame ordinal number;
(3) voice frame is calculated4 rank semi-invariant diagonal slicesWhereinFor the voice frame sampling periodIntegral multiple, and construct 4 rank semi-invariant diagonal slices vectors of the i-th frame voice frame, and do stdn and can obtain
;
(4) voice frame is estimatedFundamental tone cycle parameter, and be designated as;
(5) will,Input and output as generalized regression nerve networks GRNNs learning sample are trained, n be input and output sample to sum, its core widthForSquare root of the variance;
(6) to the temporally order framing of input speech signal sampling point, and 4 corresponding with it rank semi-invariant diagonal slices are calculated;
(7) willIt is input in the GRNNs trained, can fundamental tone cycle of this voice frame��
2. the Periodical pitch detection method under gauss heat source model environment according to claim 1, it is characterised in that, in described step (2), each frame comprises 200 voice sampling points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410124483.6A CN103903624B (en) | 2014-03-31 | 2014-03-31 | Periodical pitch detection method under a kind of gauss heat source model environment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410124483.6A CN103903624B (en) | 2014-03-31 | 2014-03-31 | Periodical pitch detection method under a kind of gauss heat source model environment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103903624A CN103903624A (en) | 2014-07-02 |
| CN103903624B true CN103903624B (en) | 2016-06-01 |
Family
ID=50994906
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410124483.6A Expired - Fee Related CN103903624B (en) | 2014-03-31 | 2014-03-31 | Periodical pitch detection method under a kind of gauss heat source model environment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103903624B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107025911B (en) * | 2016-01-29 | 2019-03-12 | 重庆工商职业学院 | Fundamental frequency detection method based on particle group optimizing |
| CN107045875B (en) * | 2016-02-03 | 2019-12-06 | 重庆工商职业学院 | fundamental tone frequency detection method based on genetic algorithm |
| CN107039051B (en) * | 2016-02-03 | 2019-11-26 | 重庆工商职业学院 | Fundamental frequency detection method based on ant group optimization |
| CN108507782B (en) * | 2018-01-29 | 2020-02-21 | 江苏大学 | Method for detecting period signal crypto period under strong background noise |
-
2014
- 2014-03-31 CN CN201410124483.6A patent/CN103903624B/en not_active Expired - Fee Related
Non-Patent Citations (5)
| Title |
|---|
| PITCH DETERMINATION OF NOISY SPEECH USING HIGHER ORDER;Asuncion Moreno et al;《Acoustics, Speech, and Signal Processing, 1992. ICASSP-92., 1992 IEEE International》;19920326;第1卷;全文 * |
| 基于三阶累积量对角切片的信号特征检测;范虹等;《计算机工程与应用》;20061231(第36期);全文 * |
| 基于切片谱和神经网络的旋转机械故障诊断方法;周鹏，秦树人;《计量技术》;20071231(第9期);全文 * |
| 基于四阶累积量对角切片的短波自适应通信信号检测;柯宏发等;《2006军事电子信息学术会议论文集》;20061231;全文 * |
| 基于对角切片谱的小波神经网络水下目标识别;顾江建等;《计算机仿真》;20120229;第29卷(第2期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103903624A (en) | 2014-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103903624B (en) | Periodical pitch detection method under a kind of gauss heat source model environment | |
| CN103559888B (en) | Based on non-negative low-rank and the sound enhancement method of sparse matrix decomposition principle | |
| CN103325381B (en) | A kind of speech separating method based on fuzzy membership functions | |
| CN111081268A (en) | Phase-correlated shared deep convolutional neural network speech enhancement method | |
| CN109378010A (en) | Training method, the speech de-noising method and device of neural network model | |
| CN108597496A (en) | A kind of speech production method and device for fighting network based on production | |
| CN106504763A (en) | Based on blind source separating and the microphone array multiple target sound enhancement method of spectrum-subtraction | |
| CN104021373A (en) | Semi-supervised speech feature variable factor decomposition method | |
| CN110164472A (en) | Noise classification method based on convolutional neural networks | |
| CN106653056A (en) | Fundamental frequency extraction model based on LSTM recurrent neural network and training method thereof | |
| Li et al. | Sams-net: A sliced attention-based neural network for music source separation | |
| CN103021405A (en) | Voice signal dynamic feature extraction method based on MUSIC and modulation spectrum filter | |
| CN107247962A (en) | A kind of real-time electrical appliance recognition and system based on sliding window | |
| CN116434759B (en) | Speaker identification method based on SRS-CL network | |
| CN106340304B (en) | A kind of online sound enhancement method under the environment suitable for nonstationary noise | |
| CN102820037B (en) | Chinese initial and final visualization method based on combination feature | |
| Xu et al. | The extraction and simulation of Mel frequency cepstrum speech parameters | |
| WO2018001125A1 (en) | Method and device for audio recognition | |
| CN105741853B (en) | A kind of digital speech perceptual hash method based on formant frequency | |
| CN102637438A (en) | Voice filtering method | |
| Hongyan et al. | Blind separation of noisy mixed speech signals based Independent Component Analysis | |
| CN105206259A (en) | Voice conversion method | |
| CN111785262B (en) | Speaker age and gender classification method based on residual error network and fusion characteristics | |
| CN103559886A (en) | Speech signal enhancing method based on group sparse low-rank expression | |
| CN104636313B (en) | A kind of redundancy extends the Blind Signal Separation method of single source observation signal |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160601 Termination date: 20190331 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |