CN202663468U - Dual-ended sounding robust structure - Google Patents
Dual-ended sounding robust structure Download PDFInfo
- Publication number
- CN202663468U CN202663468U CN 201220228850 CN201220228850U CN202663468U CN 202663468 U CN202663468 U CN 202663468U CN 201220228850 CN201220228850 CN 201220228850 CN 201220228850 U CN201220228850 U CN 201220228850U CN 202663468 U CN202663468 U CN 202663468U
- Authority
- CN
- China
- Prior art keywords
- far
- data cache
- cache module
- voice data
- 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
Landscapes
- Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
- Telephone Function (AREA)
Abstract
The utility model provides a dual-ended sounding robust structure comprising a far-end voice data cache module for receiving and storing far-end voice data, and a near-end voice data cache module for receiving and storing near-end voice data, wherein the far-end voice data cache module and the near-end voice data cache module are connected with an auto-regression model; the auto-regression model comprises a self-adapting filter capable of carrying out self-adapting study according to an evaluated error; the auto-regression model is connected with a residual echo suppression module capable of outputting processed signals to a far end; and a self-adaptive filter carries out adaptive control when double-end sounding is conducted, so that the study speed is reduced. With the adoption of the auto-regression model and a self-adaptive filter updating technology adopted by the dual-ended sounding robust structure provided by the utility model, a far-end voice acceleration convergence speed is estimated by using white noise through the auto-regression model; and meanwhile, the study speed is adaptively adjusted according to a communication scene, so that the problem of filter diverging in a dual-end sounding condition can be avoided. The voice signal quality in full-duplex communication can be ensured, and the dual-ended sounding robust structure can be widely applied to a mobile communication field.
Description
Technical field
The utility model belongs to mobile communication voice and strengthens technical field, and the device and method that particularly acoustic echo is eliminated in a kind of mobile communication system specifically, is a kind of both-end pronunciation robust structure.
Background technology
Acoustic echo refers to be transferred to again far-end after far-end speech that the near-end loud speaker plays is by the collection of near-end microphone so that remote subscriber hear own before one's voice in speech.The communication quality that had severe jamming of acoustic echo.Acoustic echo canceller is the effective way that addresses this problem.Therefore, in mobile communication system, the Echo Canceller part that is absolutely necessary.In order to make voice signal not disturbed by acoustic echo, Echo Canceller is in the situation of single-ended pronunciation (single talk, ST), and the terminal coupling loss of weighting will reach 46dB at least; And in the situation of both-end pronunciation (double talk, DT), be greater than 26dB.The principle of Echo Canceller be with far-end speech as the reference signal, approach coupling echo path from the loud speaker to the microphone by sef-adapting filter, the echo signal that obtains being similar to, and near end signal, deduct the echo of this part estimation.Desirable Echo Canceller should have Fast Convergent and tracking (time-varying characteristics of reply echo path) ability under the prerequisite that guarantees low mismatch value (misalignment).These two characteristics all depend on the adaptive algorithm that filter adopts.In the ST situation, existing adaptive algorithm can both obtain satisfied echo eradicating efficacy.In the DT situation, owing to being subjected to the interference of near-end speech, these adaptive algorithms will be dispersed, and depart from actual echo channel, so that the performance that echo is eliminated sharply worsens.According to statistics, in communication process, the time that DT occurs accounts for about 20%.Therefore the performance of Echo Canceller is very important during DT.Solution commonly used is to introduce both-end utterance detection device (double talk detector, DTD).When DTD detected the both-end pronunciation, sef-adapting filter stopped to upgrade.The shortcoming of this solution is that filter may be dispersed when DTD detects the both-end pronunciation.For this problem, people have proposed various variable step size adaptive algorithms.Its thinking is according to characteristics of speech sounds far away, near-end, the renewal step-length of automatically regulating sef-adapting filter, and step-length is larger during ST, and step-length is less during DT.But these variable step size methods convergence rate when ST and echo path change can be affected.
Summary of the invention
Technical problem to be solved in the utility model is the present situation for prior art, thereby the voice quality in the quick follow-up control assurance full-duplex communication is provided in the situation that provides a kind of single-ended pronunciation and echo path to change, has avoided a kind of both-end pronunciation robust structure of filter divergence problem in the both-end pronunciation situation.
The utility model solves the problems of the technologies described above the technical scheme that adopts:
A kind of both-end pronunciation robust structure, include the near-end speech data cache module for the far-end speech data cache module that receives and preserve the far-end speech data and reception and preservation near-end speech data, wherein, far-end speech data cache module and near-end speech data cache module signal are connected with autoregression model, autoregression model includes the sef-adapting filter that can carry out according to evaluated error adaptive learning, and autoregression model is connected with the residual echo that signal after processing can be exported to far-end and suppresses module.
For optimizing technique scheme, Adopts measure also comprises:
Above-mentioned residual echo suppresses module and is connected with the equalization processing device.
Above-mentioned residual echo suppresses module and is connected with the automatic gain control processor.
Compared with prior art, a kind of both-end pronunciation of the utility model robust structure, include the near-end speech data cache module for the far-end speech data cache module that receives and preserve the far-end speech data and reception and preservation near-end speech data, wherein, far-end speech data cache module and near-end speech data cache module signal are connected with autoregression model, autoregression model includes the sef-adapting filter that can carry out according to evaluated error adaptive learning, and autoregression model is connected with the residual echo that signal after processing can be exported to far-end and suppresses module; Sef-adapting filter carries out self adaptation when both-end pronounces regulates, and reduces pace of learning.The sef-adapting filter pace of learning is very fast in the situation that single-ended pronunciation or echo path change, and pace of learning is slower in the both-end pronunciation situation.Owing to having adopted autoregression model and sef-adapting filter renewal technology, but estimate near-end speech convergence speedup speed with white noise by autoregression model, regulate the problem that pace of learning has been avoided filter divergence in the both-end pronunciation situation according to the communication scenes self adaptation simultaneously.Sef-adapting filter only can be eliminated linear echo, for nonlinear echo, also needs to suppress through residual echo the processing of module.The utility model can guarantee the quality of speech signal in the full-duplex communication, can be widely used in the moving communicating field.
Description of drawings
Fig. 1 is the utility model both-end pronunciation robust structure schematic diagram;
Fig. 2 is the schematic diagram that the utility model acoustic echo is eliminated.
Embodiment
Embodiment is described in further detail the utility model below in conjunction with accompanying drawing.
To shown in Figure 2, figure grade is described as follows far-end speech data cache module 1, near-end speech data cache module 2, autoregression model 3, sef-adapting filter 4, residual echo inhibition module 5, equalization processing device 6, automatic gain control processor 7 such as Fig. 1.
Fig. 1 is to a kind of both-end pronunciation robust structure of the present utility model shown in Figure 2, include the near-end speech data cache module 2 for the far-end speech data cache module 1 that receives and preserve the far-end speech data and reception and preservation near-end speech data, wherein, far-end speech data cache module 1 and near-end speech data cache module 2 signals are connected with autoregression model 3, autoregression model 3 includes the sef-adapting filter 4 that can carry out according to evaluated error adaptive learning, and autoregression model 3 is connected with the residual echo that signal after processing can be exported to far-end and suppresses module 5; Sef-adapting filter 4 carries out self adaptation when both-end pronounces regulates, and reduces pace of learning.
Among the embodiment, residual echo suppresses module 5 and is connected with equalization processing device 6.
Among the embodiment, residual echo suppresses module 5 and is connected with automatic gain control processor 7.
A kind of method of eliminating acoustic echo by both-end pronunciation robust of the present utility model may further comprise the steps:
Step 1: estimate reference signal as echo after with the far-end speech data buffer storage by far-end speech data cache module 1;
Step 2: estimate reference signal as Mike's voice after with near-end speech data buffer memory by near-end speech data cache module 2;
Step 3: estimate reference signal according to near-end speech, estimate near-end voice signals with autoregression model 3, the coefficient of autoregression model 3 carries out adaptive learning according to evaluated error;
Step 4: estimate reference signal according to echo, estimate to be coupled to from loud speaker Mike's echo by sef-adapting filter 4, and regulate the regularization factor according to the energy self-adaptation of residual echo, near-end speech and far-end speech, adjust sef-adapting filter 4 pace of learnings;
Step 5: the echo that near-end speech is deducted estimation obtains error signal;
Step 6: the error signal that step 5 is obtained suppresses the input of module 5 as residual echo;
Step 7: after residual echo being suppressed the processing of output through equalization processing device 6, automatic gain control processor 7 of module 5, send to far-end.
Among the embodiment, sef-adapting filter 4 is auto-adaptive fir filter.
Among the embodiment, the pace of learning of sef-adapting filter 4 when single-ended pronunciation or echo path change is higher than the pace of learning when both-end pronounces.
Among the embodiment, residual echo suppresses module 5 according to communications status, and near-end and far-end speech are decayed.
Sef-adapting filter 4 pace of learnings are very fast in the situation that single-ended pronunciation or echo path change, and pace of learning is slower in the both-end pronunciation situation.Owing to having adopted autoregression model 3 and sef-adapting filter 4 renewal technology, but estimate near-end speech convergence speedup speed with white noise by autoregression model 3, regulate the problem that pace of learning has been avoided filter divergence in the both-end pronunciation situation according to the communication scenes self adaptation simultaneously.Sef-adapting filter 4 only can be eliminated linear echo, for nonlinear echo, also needs to suppress through residual echo the processing of module 5.The utility model can guarantee the quality of speech signal in the full-duplex communication, can be widely used in the moving communicating field.
Main design of the present utility model is to utilize autoregression model 3 with near-end speech and far-end speech decorrelation, to reach the purpose of convergence speedup speed; Adjust simultaneously the pace of learning of sef-adapting filter 4 according to the energy statistics characteristic of near, remote signaling and residual echo, under guaranteeing single-ended pronunciation and echo path change situation in the convergence rate, avoided dispersing of filter in the both-end pronunciation situation.
Above embodiment is described preferred implementation of the present utility model; be not that scope of the present utility model is limited; relate under the prerequisite of spirit not breaking away from the utility model; various distortion and improvement that the common engineers and technicians in this area make the technical solution of the utility model all should fall in the definite protection range of claims of the present utility model.
Claims (3)
1. both-end pronunciation robust structure, include the near-end speech data cache module (2) for the far-end speech data cache module (1) that receives and preserve the far-end speech data and reception and preservation near-end speech data, it is characterized in that: described far-end speech data cache module (1) and near-end speech data cache module (2) signal are connected with autoregression model (3), described autoregression model (3) includes the sef-adapting filter (4) that can carry out according to evaluated error adaptive learning, and described autoregression model (3) is connected with the residual echo that signal after processing can be exported to far-end and suppresses module (5).
2. according to a kind of both-end pronunciation robust structure according to claim 1, it is characterized in that: described residual echo suppresses module (5) and is connected with equalization processing device (6).
3. according to a kind of both-end pronunciation robust structure according to claim 2, it is characterized in that: described residual echo suppresses module (5) and is connected with automatic gain control processor (7).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201220228850 CN202663468U (en) | 2012-05-21 | 2012-05-21 | Dual-ended sounding robust structure |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201220228850 CN202663468U (en) | 2012-05-21 | 2012-05-21 | Dual-ended sounding robust structure |
Publications (1)
Publication Number | Publication Date |
---|---|
CN202663468U true CN202663468U (en) | 2013-01-09 |
Family
ID=47458340
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 201220228850 Expired - Fee Related CN202663468U (en) | 2012-05-21 | 2012-05-21 | Dual-ended sounding robust structure |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN202663468U (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102655558A (en) * | 2012-05-21 | 2012-09-05 | 宁波工程学院 | Double-end pronouncing robust structure and acoustic echo cancellation method |
-
2012
- 2012-05-21 CN CN 201220228850 patent/CN202663468U/en not_active Expired - Fee Related
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102655558A (en) * | 2012-05-21 | 2012-09-05 | 宁波工程学院 | Double-end pronouncing robust structure and acoustic echo cancellation method |
CN102655558B (en) * | 2012-05-21 | 2013-10-09 | 宁波工程学院 | Double-end pronouncing robust structure and acoustic echo cancellation method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US9589556B2 (en) | Energy adjustment of acoustic echo replica signal for speech enhancement | |
CN105825864B (en) | Both-end based on zero-crossing rate index is spoken detection and echo cancel method | |
KR101353686B1 (en) | Communication headset speech enhancement method and device, and noise reduction communication headset | |
CN105612576B (en) | Limitation active noise eliminates output | |
CN105577961B (en) | Automatic tuning of gain controller | |
US9264807B2 (en) | Multichannel acoustic echo reduction | |
CN109716743B (en) | Full duplex voice communication system and method | |
US9343073B1 (en) | Robust noise suppression system in adverse echo conditions | |
CN107026950B (en) | A kind of frequency domain adaptive echo cancel method | |
CN1953060A (en) | Echo elimination device for microphone and method thereof | |
CN104050971A (en) | Acoustic echo mitigating apparatus and method, audio processing apparatus, and voice communication terminal | |
US10348887B2 (en) | Double talk detection for echo suppression in power domain | |
CN107396223B (en) | Earphone uplink noise-reduction method | |
CN111145771A (en) | Voice signal processing method, processing device, terminal and storage medium thereof | |
US9508359B2 (en) | Acoustic echo preprocessing for speech enhancement | |
CN107134281A (en) | Adaptive filter coefficient update method during a kind of adaptive echo is eliminated | |
CN110956975A (en) | Echo cancellation method and device | |
CN106657700B (en) | hand-free talking device capable of eliminating echo and its control method | |
CN107071196B (en) | A kind of adaptive echo cancellation method | |
CN112492112A (en) | Echo eliminating method and device based on intercom system | |
CN102655558B (en) | Double-end pronouncing robust structure and acoustic echo cancellation method | |
WO2015044915A1 (en) | Acoustic feedback cancellation based on cesptral analysis | |
US9779753B2 (en) | Method and apparatus for attenuating undesired content in an audio signal | |
CN202663468U (en) | Dual-ended sounding robust structure | |
CA2429877C (en) | Echo canceller employing h-register and storage register |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
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: 20130109 Termination date: 20150521 |
|
EXPY | Termination of patent right or utility model |