WO2014056328A1 - 一种回声抵消方法和设备 - Google Patents
一种回声抵消方法和设备 Download PDFInfo
- Publication number
- WO2014056328A1 WO2014056328A1 PCT/CN2013/076685 CN2013076685W WO2014056328A1 WO 2014056328 A1 WO2014056328 A1 WO 2014056328A1 CN 2013076685 W CN2013076685 W CN 2013076685W WO 2014056328 A1 WO2014056328 A1 WO 2014056328A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- band
- signal
- echo
- echo cancellation
- frequency domain
- 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.)
- Ceased
Links
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10K—SOUND-PRODUCING DEVICES; METHODS OR DEVICES FOR PROTECTING AGAINST, OR FOR DAMPING, NOISE OR OTHER ACOUSTIC WAVES IN GENERAL; ACOUSTICS NOT OTHERWISE PROVIDED FOR
- G10K11/00—Methods or devices for transmitting, conducting or directing sound in general; Methods or devices for protecting against, or for damping, noise or other acoustic waves in general
- G10K11/16—Methods or devices for protecting against, or for damping, noise or other acoustic waves in general
- G10K11/175—Methods or devices for protecting against, or for damping, noise or other acoustic waves in general using interference effects; Masking sound
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M9/00—Arrangements for interconnection not involving centralised switching
- H04M9/08—Two-way loud-speaking telephone systems with means for conditioning the signal, e.g. for suppressing echoes for one or both directions of traffic
- H04M9/082—Two-way loud-speaking telephone systems with means for conditioning the signal, e.g. for suppressing echoes for one or both directions of traffic using echo cancellers
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10K—SOUND-PRODUCING DEVICES; METHODS OR DEVICES FOR PROTECTING AGAINST, OR FOR DAMPING, NOISE OR OTHER ACOUSTIC WAVES IN GENERAL; ACOUSTICS NOT OTHERWISE PROVIDED FOR
- G10K2210/00—Details of active noise control [ANC] covered by G10K11/178 but not provided for in any of its subgroups
- G10K2210/50—Miscellaneous
- G10K2210/505—Echo cancellation, e.g. multipath-, ghost- or reverberation-cancellation
Definitions
- FIG. 1 is a schematic diagram of an application scenario of an echo cancellation technology.
- a remote signal transmitted from a network is decoded by a CODEC, that is, after being decoded by a codec module, and then sent to a speaker for playback, and the microphone not only picks up the near-end signal, but also simultaneously
- the far-end signal (ie echo) played by the speaker is also picked up.
- the signal picked up by the microphone is echo-cancelled, the echo in the microphone signal is cancelled, leaving only the near-end signal, which is sent through the network after being encoded by the CODEC module. remote. If the echo in the microphone signal is not removed, the far end will hear its own sound when it is sent to the far end.
- the purpose of the echo cancellation technique is to eliminate the echo as much as possible while retaining the near-end signal.
- Echo cancellation processing usually consists of two parts: AEC (Adaptive Echo Canceller), adaptive echo cancellation, and RES (Residual Echo Suppressor), residual echo suppression.
- AEC Adaptive Echo Canceller
- RES Residual Echo Suppressor
- Adaptive echo cancellation uses an adaptive filter to simulate the spatial echo path and eliminate the echo in the microphone signal.
- the AEC module cannot eliminate the echo very cleanly, so the RES module is required to be residual.
- the echo performs further echo suppression processing.
- the adaptive filtering has algorithms such as NLMS, RLS, and MDF (Multidelay block frequency domain adaptive filter), which is an implementation form of the block NLMS algorithm in the frequency domain.
- NLMS Long reverberation time
- RLS Low-latency L1
- MDF Multidelay block frequency domain adaptive filter
- the adaptive filter is computationally intensive, thereby increasing the cost of the device.
- the invention provides a low complexity The echo cancellation algorithm.
- the prior art uses sub-band adaptive filtering to solve this problem. As shown
- the near-end signal " and the far-end signal are respectively sub-band divided, each sub-band has a bandwidth of 250 Hz, so there are 16 sub-bands in 8KHz, 32 sub-bands in 16KHz, and 64 sub-bands in 32KHz.
- Each sub-band uses the NLMS algorithm for echo cancellation, and then each sub-band is integrated to obtain a residual echo signal.
- An embodiment of the present invention provides an echo cancellation method, where the method includes:
- the AEC module performs adaptive filtering echo cancellation processing on the low-band audio signal to be processed
- the RES module performs envelope prediction echo suppression on the high-band audio signal to be processed, echo suppression on the low-band audio signal to be processed, and outputs the processed signal.
- the method effectively reduces the computational complexity of echo cancellation and solves the problem of echo leakage at the subband boundary.
- the embodiment of the invention simultaneously provides an echo cancellation AEC module, including:
- the reference signal analysis filter processes the reference signal into a high-band reference signal and a low-band reference signal, outputs the high-band reference signal to the artificial echo synthesis filter, and outputs the low-band reference signal to the adaptive filter;
- the audio signal analysis filter to be processed processes the audio signal to be processed into a high-band audio signal to be processed and a low-band audio signal to be processed, and outputs the high-band audio signal to be processed to the preliminary echo cancellation signal.
- the adaptive filter is connected to the reference signal analysis filter and the to-be-processed audio signal analysis filter, and the low-band reference signal is subjected to adaptive filtering processing to generate a low-band artificial echo signal, and output to the artificial echo synthesis filter;
- the low-band to-be-processed audio signal is subjected to adaptive filtering processing to generate a low-band preliminary echo cancellation signal, and is output to the preliminary echo cancellation signal synthesis filter;
- the artificial echo synthesis filter is connected to the adaptive filter, and combines the high-band reference signal and the low-band artificial echo signal into an artificial echo signal, and outputs the signal to the RES device;
- the preliminary echo cancellation signal synthesis filter is connected to the adaptive filter, and combines the high-band to-be-processed audio signal and the low-band preliminary echo cancellation signal into a preliminary echo cancellation signal, and outputs it to the RES module.
- An embodiment of the present invention further provides an echo cancellation RES module, including:
- the FFT transform module performs FFT transformation on the reference signal, the to-be-processed audio signal, the artificial echo signal, and the preliminary echo cancellation signal to generate a corresponding reference frequency domain signal, a frequency domain audio signal to be processed, an artificial echo frequency domain signal, and a preliminary echo cancellation frequency.
- the domain signal; and the frequency domain audio signal to be processed, the artificial echo frequency domain signal, and the preliminary echo cancellation frequency domain signal are input to the RES low band module, and the reference frequency domain signal and the preliminary echo cancellation frequency domain signal are input to the RES high band module;
- the RE S high-band module performs envelope prediction echo suppression on the high-band signal in the preliminary echo cancellation frequency domain signal, and calculates and outputs the residual echo suppression coefficient;
- the RES low-band module performs echo suppression on the low-band signal in the preliminary echo cancellation frequency domain signal, and outputs the processing result.
- the device effectively reduces the computational complexity of echo cancellation and solves the problem of echo leakage at the subband boundary.
- FIG. 3 is a structural diagram of a device according to Embodiment 1 of the present invention.
- FIG. 5 is a flowchart of a method according to Embodiment 1 of the present invention.
- FIG. 7 is a reference diagram of a sub-band division manner according to Embodiment 1 of the present invention.
- FIG. 8 is a diagram showing an example of an echo impulse response and an energy attenuation curve according to Embodiment 1 of the present invention.
- FIG. 9 is a flowchart of QMF analysis and synthesis filter processing according to Embodiment 1 of the present invention. detailed description
- Embodiments of the present invention provide an echo cancellation method, which cancels an echo in an audio signal to be processed, and outputs a signal after echo cancellation.
- reference signal; ) ie, the far-end signal, the reference signal and the far-end signal have the same meaning in this patent
- the microphone picks up the to-be-processed audio signal including the echo signal _y( «)
- the local signal and the noise to be processed audio signal are processed by the AEC module and the RES module to eliminate the echo signal ⁇ ) and noise, and retain the local signal.
- the method includes the following steps, please refer to FIG. 4:
- the AEC module performs preliminary echo and gram processing on the audio signal to be processed, and generates a preliminary echo ⁇ cancellation signal
- the AEC module processes the reference signal x( «), the audio signal to be processed “), outputs the initial echo cancellation signal artificial echo signal ( «) and the low-band echo filter coefficient H(t) in the frequency domain; and the reference signal ⁇ ( «) and the pending audio signal ”) are output to the RES module. Specifically, it includes:
- the reference signal; ) is processed by the QMF analysis filter to generate a high-band reference signal x hi (n) and a low-band reference signal w ( «), wherein the high-band reference signal is output to the QMF synthesis filter, and the low-band reference signal is output w ( «) to the adaptive filter;
- the low-band reference signal w ( «) is processed by an adaptive filter to generate a low-band artificial echo signal y l ' ow (n) and output to the QMF synthesis filter;
- the to-be-processed audio signal is generated, and a high-band audio signal to be processed and a low-band audio signal to be processed are generated, and the high-band audio signal d u (n) to QMF synthesis filter is outputted.
- the low-band pending audio signal d low ( «) minus the low-band artificial echo signal ⁇ n) to obtain a low-band preliminary echo cancellation signal e to ) and output to the QMF synthesis filter module;
- the high-band pending audio signal O) and the low-band preliminary echo cancellation signal are processed by the QMF synthesis filter to generate a preliminary echo cancellation signal and output to the RES module;
- the low-band echo filter coefficient in the frequency domain is calculated by the process of adaptive filtering to process the low-band reference signal
- the AEC module also outputs a reference signal x( «) and an audio signal to be processed to the RES module as an auxiliary processing signal for further echo cancellation by the RES module.
- S2 The RES module performs further echo cancellation processing on the preliminary echo cancellation signal e(n) to generate an echo cancellation signal, please refer to FIG. 5.
- S201 The RES module performs fast Fourier transform (FFT) on the reference signal x( «), the artificial echo signal (") of the audio signal to be processed, and the preliminary echo cancellation signal, respectively, to generate a corresponding reference frequency domain signal Jr.
- FFT fast Fourier transform
- the RES high-band module performs envelope prediction echo suppression on the high-band signal (hereinafter referred to as high-band signal) in the preliminary echo cancellation frequency domain signal £), calculates and outputs the residual echo suppression coefficient G he (k); RES low band
- the module performs echo suppression on the low-band signal (hereinafter referred to as low-band signal) in the initial echo cancellation frequency domain signal by using the existing echo suppression technology, and outputs the processing result; please refer to FIG. 6.
- the low band RES performs residual echo suppression on the lines below 263, and the high band pair 231 The line above the root is suppressed, and there are a total of 32 lines overlapping.
- the RES high-band module calculates the residual echo suppression coefficient as follows: according to the initial echo cancellation frequency domain signal, each sub-band energy £( ), the reference frequency domain signal JT( t), each sub-band energy f/(0, residual back Each sub-band energy of the audio domain signal, the high-band envelope prediction coefficient gl (0 and the attenuation factor factor of the high-band envelope prediction coefficient are calculated for each sub-band gain; the high-band spectral line gain G he (k is calculated according to each sub-band gain) ), that is, the residual echo coefficient. Specifically, it includes:
- the high-band RES module uses the envelope prediction method to estimate the echo of the high band and then suppresses it.
- the so-called envelope prediction is to consider the influence of the phase.
- i denotes the subband number divided in the frequency domain
- L is the number of frames used for envelope prediction.
- ⁇ () indicates the first coefficient of the i-th sub-band, f/_, (0 indicates the energy of the i-th sub-band of the first frame before the reference frequency domain signal.
- the number of spectral lines is 512
- the sub-band with high RES Please refer to Figure 7 for the division method of the bands. There are 5 sub-bands in total, and the sub-bands are overlapped by means of triangular windows.
- the sub-bands are overlapped by means of a triangular window.
- the present invention is not limited to this method, and may be overlapped by a sine window or a rectangular window.
- the audio signal to be processed includes the local signal of the echo signal and the noise.
- the initial echo cancellation signal is generated by the FFT transform to generate a preliminary echo cancellation frequency domain signal. Therefore, the initial echo 4 frequency cancellation domain signal E ⁇ J signal also includes three Part: Residual echo frequency domain signal, local frequency domain signal and noise frequency domain signal.
- w is the starting frequency of the i-th sub-band
- (0 is the end frequency of the i-th sub-band, which is the weighting coefficient of each frequency.
- ⁇ (0 is the center frequency of the i-th sub-band
- ⁇ () is as follows Formula value:
- ⁇ is the frame length. Since only the low-band echo filter coefficient H(A) in the frequency domain can be obtained, the low-band echo filter coefficient can only be used to approximate the high-band filter coefficient.
- the specific implementation method is to perform IFFT transformation on the HW to obtain the time domain. The low-band impulse response is then high-pass filtered to obtain the impulse response A 0 canal of the high sub-band, and then the envelope prediction coefficient is calculated using the above formula:
- each subband of the high band uses the same envelope prediction coefficient t
- FIG 8 is an example of an echo impulse response and an energy decay curve (EDC):
- EDC is substantially linear in the middle portion.
- EDC is defined as:
- the attenuation factor of the envelope prediction coefficient can be calculated.
- the slope of the EDC can be calculated by a linear recursive method. It can be seen that the EDC is linear except for the initial tens of milliseconds belonging to the early echo, that is, the attenuation factor of the envelope prediction coefficient is a fixed value.
- the number of frames used to estimate the echo by the envelope prediction coefficient is relatively large. For example, the echo tail is 256 ms, and the length of each frame is 8 ms, which requires 32 frames, which requires more storage space and calculation amount, and uses the attenuation factor. It is a fixed value feature that can reduce the storage space and the amount of calculation.
- the specific method is as follows:
- the DB difference of ⁇ E db (j)-E db (i) is proportional.
- the gain of each band of the high band that is, the residual echo suppression coefficient can be obtained by adjacent subband gain interpolation:
- G(k) w k (i)G(i) + w k (il)G(il) (17) w k () is the coefficient of the previous calculation of the energy of each sub-band.
- S207 Perform an inverse fast Fourier Transformation (IFFT) on the echo canceled signal £' (A), then backfill the noise, and output the final processed signal.
- IFFT inverse fast Fourier Transformation
- the backfill noise module determines whether the current frame is a residual echo according to the NLP identifier, and if so, backfills the noise, otherwise determines the VAD flag, and if the current frame is noise, saves the noise signal to the noise buffer.
- the signal is divided into a high-band signal and a low-band signal, and the existing echo cancellation algorithm is applied to the low-band signal, and the echo suppression algorithm for the envelope prediction of the high-band signal is used to greatly reduce the computational complexity.
- the effect of the degree can avoid the problem of subband adaptive filtering boundary leakage.
- Embodiments of the present invention simultaneously provide an echo cancellation AEC module, including
- the reference signal analysis filter processes the reference signal into a high-band reference signal and a low-band reference signal, outputs the high-band reference signal to the artificial echo synthesis filter, and outputs the low-band reference signal to the adaptive filter;
- the audio signal analysis filter to be processed the audio signal to be processed is processed into a high-band audio signal to be processed and a low-band audio signal to be processed, and the high-band audio signal to be processed is output to the preliminary echo cancellation signal synthesis filter; Processing the audio signal output to the adaptive filter;
- the low-band reference signal is subjected to adaptive filtering to generate a low-band artificial echo signal, and output to the artificial echo synthesis filter;
- the low-band pending audio signal is subjected to adaptive filtering to generate a low-band preliminary echo cancellation signal, and output To the preliminary echo cancellation signal synthesis filter;
- the artificial echo synthesis filter is connected to the adaptive filter, and combines the high-band reference signal and the low-band artificial echo signal into an artificial echo signal, and outputs the signal to the RES device;
- the preliminary echo cancellation signal synthesis filter is connected to the adaptive filter, and combines the high-band to-be-processed audio signal and the low-band preliminary echo cancellation signal into a preliminary echo cancellation signal, and outputs it to the RES module.
- the QMF analysis and synthesis filter processing flow is shown in Figure 9.
- the sample is doubled.
- the low-band processing and the high-band processing are respectively performed, and then the double-up sample is separately performed, and the integrated filters F0(z) and Fl(z) are filtered, and the combined signal y(n) is output after the addition.
- the filter has the following limitations:
- the embodiment of the invention also provides an echo cancellation RES module, including:
- the FFT transform module performs FFT transformation on the reference signal, the to-be-processed audio signal, the artificial echo signal, and the preliminary echo cancellation signal to generate a corresponding reference frequency domain signal, a frequency domain audio signal to be processed, an artificial echo frequency domain signal, and a preliminary echo cancellation frequency. Domain signal; and will process the frequency domain audio signal
- the number, the artificial echo frequency domain signal, and the preliminary echo cancellation frequency domain signal are input to the RES low band module, and the reference frequency domain signal and the preliminary echo cancellation frequency domain signal are input to the RES high band module;
- the RE S high-band module performs envelope prediction echo suppression on the high-band signal in the preliminary echo cancellation frequency domain signal, and calculates and outputs the residual echo suppression coefficient;
- the RES low-band module performs echo suppression on the low-band signal in the preliminary echo cancellation frequency domain signal, and outputs the processing result.
- the RES high band module includes:
- the preliminary echo cancels each sub-band energy calculation module of the frequency domain signal, calculates a preliminary echo cancellation frequency band sub-band energy, and outputs the preliminary echo cancellation frequency domain signal sub-band energy to the sub-band gain calculation module;
- a reference frequency domain signal sub-band energy calculation module configured to calculate each sub-band energy of the reference frequency domain signal, and output the sub-band energy of the reference frequency domain signal to the sub-band gain calculation module;
- An IFFT transform module performs IFFT transform on the low-band echo filter coefficients to generate a low-band impulse response in the time domain, and outputs the impulse response to the high-pass filter module;
- the high-pass filtering module is connected to the IFFT conversion module, performs high-pass filtering processing on the low-band impulse response of the time domain, generates a high-band impulse response, and outputs the high-band impulse response to High band envelope prediction coefficient calculation module;
- the high-band envelope prediction coefficient calculation module is connected to the high-pass filter module, calculates a high-band envelope prediction coefficient according to the high-band impulse response, and outputs the high-band envelope prediction coefficient to an EDC slope a calculation module and the sub-band gain calculation module;
- the EDC slope calculation module is connected to the high-band envelope prediction coefficient calculation module, calculates an EDC slope, and outputs the EDC slope to an attenuation factor calculation module;
- the attenuation factor calculation module is connected to the EDC slope calculation module, calculates an attenuation factor according to the EDC slope, and outputs the attenuation factor to the sub-band gain calculation module;
- the subband gain calculation module is connected to each subband energy calculation module of the preliminary echo cancellation frequency domain signal, each subband energy calculation module of the reference frequency domain signal, a high band envelope prediction coefficient calculation module, and an attenuation factor calculation module. Accomplishing the energy of each sub-band of the frequency domain signal according to the preliminary echo, reference Frequency sub-band energy, high-band envelope prediction coefficient, and attenuation factor are used to calculate sub-band gain, and the sub-band gain is output to a high-band spectral line gain calculation module;
- the high-band spectral line gain calculation module is connected to the sub-band gain calculation module, and calculates a high-band spectral line gain, that is, a residual echo suppression coefficient, according to the sub-band gain.
- the device provided by the embodiment of the invention divides the signal into a high-band signal and a low-band signal, and uses the existing echo cancellation algorithm for the low-band signal, and the echo suppression algorithm for the high-band signal envelope prediction. Reduce the computational complexity and avoid the problem of subband adaptive filtering boundary leakage.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
- Telephone Function (AREA)
Abstract
一种回声抵消方法和设备,该方法包括:划分待处理音频信号为高带待处理音频信号和低带待处理音频信号;AEC模块对低带待处理音频信号进行自适应滤波回声抵消处理,对高带待处理音频信号不进行自适应滤波回声抵消处理,生成初步回声抵消信号;RES模块对初步回声抵消频率信号中的高带信号进行包络预测回声抑制,计算并输出残留回声抑制系数;RES模块对初步回声抵消频域信号中的低带信号进行回声抑制,输出处理结果;将输出结果与残留回声抑制系数相乘输出回声抵消后信号。该设备包括AEC模块和RES模块。所述的方法和设备通过对待处理信号分别进行高带和低带处理,既降低了回声抵消的计算量,又解决了子带边界存在回声泄露的问题。
Description
一种回声抵消方法和设备 本申请要求于 2012年 10月 12 日提交中国专利局、 申请号为
201210387313.8、 发明名称为 "一种回声抵消方法和设备" 的中国专利申请 的优先权, 其全部内容通过引用结合在本申请中。 技术领域 本发明涉及音频信号处理领域, 尤其涉及一种回声抵消方法和设备。 背景技术 图 1是回声抵消技术的应用场景示意图,从网络上传来的远端信号经过 CODEC , 即编译码器模块解码之后, 进行模数转换, 送到扬声器播放, 麦 克风不仅拾取近端信号, 同时也拾取了扬声器播放的远端信号 (即回声) , 麦克风拾取的信号进行回声抵消处理之后, 麦克风信号中的回声被消除, 只 剩下近段信号, 经过 CODEC模块编码处理之后, 通过网络送到远端。 如果 麦克风信号中的回声没有被消除, 送到远端后, 远端会听到自己的声音, 回 声抵消技术的目的是尽量消除回声, 而保留近端信号。
回声抵消处理通常包括两个部分: AEC ( Adaptive Echo Canceller ) , 自 适应回声抵消以及 RES ( Residual Echo Suppressor ) , 残留回声抑制。 自适 应回声抵消釆用自适应滤波器模拟空间回声路径, 消除麦克风信号中的回 声, 一般情况下, 由于噪声等因素的影响, AEC 模块不能将回声消除的很 干净, 因此需要 RES模块对残留的回声进行进一步的回声抑制处理。
自适应滤波有 NLMS、 RLS、 MDF( Multidelay block frequency domain adaptive filter )算法是块 NLMS算法在频域的一种实现形式)等算法。 在混 响时间较长、 釆样率较高时, 自适应滤波器需要很长的阶数, 例如混响为 300ms , 釆样率为 48khz时, 需要的阶数为 48000 0.3= 14400, 这样导致自 适应滤波器的计算量很大, 从而增加设备的成本。 本发明提供一种低复杂度
的回声抵消算法。
现有技术为了降低复杂度, 釆用子带自适应滤波来解决这个问题。 如图
2 所示, 近端信号 《)和远端信号 分别进行子带分割, 每个子带带宽 250Hz, 因此 8KHz共有 16各子带, 16KHz共有 32个子带, 32KHz共有 64 个子带。 每个子带釆用 NLMS算法进行回声抵消, 之后各个子带进行综合, 得到残留回声信号。
但是, 现有技术至少存在以下问题:
子带边界存在回声泄露。 经调试发现子带的自适应滤波器, 在子带边界 处回声衰减量明显不足, 经常会残留一些较强的单频信号(听起来就是类似 "叽咣" 的声音, 比较影响主观感受) 。
发明内容
鉴于此, 有必要提供一种回声抵消方法和设备来解决上述问题。
本发明实施例提供了一种回声抵消方法, 该方法包括:
划分待处理音频信号为高带待处理音频信号和低带待处理音频信号;
AEC模块对低带待处理音频信号进行自适应滤波回声抵消处理;
RES模块对高带待处理音频信号进行包络预测回声抑制,对低带待处理 音频信号进行回声抑制, 输出处理后信号。
本方法既有效的降低了回声抵消的计算量,又解决了子带边界存在的回 声泄露问题。
本发明实施例同时提供一种回声抵消 AEC模块, 包括:
参考信号分析滤波器, 将参考信号处理为高带参考信号和低带参考信 号, 将高带参考信号输出到人造回声合成滤波器, 将低带参考信号输出到自 适应滤波器;
待处理音频信号分析滤波器,将待处理音频信号处理为高带待处理音频 信号和低带待处理音频信号,将高带待处理音频信号输出到初步回声抵消信
号合成滤波器; 将低带待处理音频信号输出到自适应滤波器;
所述自适应滤波器,与参考信号分析滤波器和待处理音频信号分析滤波 器相连, 将低带参考信号经过自适应滤波处理生成低带人造回声信号, 并输 出到人造回声合成滤波器;将低带待处理音频信号经过自适应滤波处理生成 低带初步回声抵消信号, 并输出到初步回声抵消信号合成滤波器;
人造回声合成滤波器, 与自适应滤波器相连, 将高带参考信号和低带人 造回声信号合成为人造回声信号, 并输出给 RES设备;
初步回声抵消信号合成滤波器, 与自适应滤波器相连, 将高带待处理音 频信号和低带初步回声抵消信号合成为初步回声抵消信号, 并输出给 RES 模块。
本发明实施例还提供一种回声抵消 RES模块, 包括:
FFT变换模块, 将参考信号、 待处理音频信号、 人造回声信号、 初步回 声抵消信号经过 FFT 变换, 生成对应的参考频域信号、 待处理频域音频信 号、 人造回声频域信号以及初步回声抵消频域信号; 并将待处理频域音频信 号、 人造回声频域信号以及初步回声抵消频域信号输入到 RES低带模块, 将参考频域信号、 初步回声抵消频域信号输入到 RES高带模块;
RE S高带模块,对初步回声抵消频域信号中的高带信号进行包络预测回 声抑制, 计算并输出残留回声抑制系数;
RES低带模块, 对初步回声抵消频域信号中的低带信号进行回声抑制, 输出处理结果。
本设备既有效的降低了回声抵消的计算量,又解决了子带边界存在的回 声泄露问题。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对实 施例或现有技术描述中所需要使用的附图作一简单地介绍, 显而易见地, 下
面描述中的附图是本发明的一些实施例, 对于本领域普通技术人员来讲, 在 不付出创造性劳动性的前提下, 还可以根据这些附图获得其他的附图。
图 1是本发明背景技术的方法流程图;
图 2是本发明背景技术中现有技术的方法流程图;
图 3是本发明实施例一的设备结构图;
图 4是本发明实施例一的方法流程图;
图 5是本发明实施例一的方法流程图;
图 6是本发明实施例一的方法流程图;
图 7是本发明实施例一的子带划分方式参考图
图 8是本发明实施例一的回声冲激响应以及能量衰减曲线的示例图; 图 9是本发明实施例一的 QMF分析和综合滤波器处理流程图。 具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚, 下面将结合本发 明实施例中的附图, 对本发明实施例中的技术方案进行清楚、 完整地描述, 显然, 所描述的实施例是本发明一部分实施例, 而不是全部的实施例。 基于 本发明中的实施例,本领域普通技术人员在没有作出创造性劳动前提下所获 得的所有其他实施例, 都属于本发明保护的范围。
本发明实施例提供一种回声抵消方法, 将待处理音频信号中的回声抵 消, 输出回声抵消后信号。
请参考图 3 , 参考信号; ) (即远端信号, 在本专利中参考信号和远端 信号具有相同含义) 输出到扬声器播放; 麦克风拾取待处理音频信号 其中包括回声信号 _y(«)、 本端信号 以及噪声 待处理音频信号经过 AEC模块和 RES模块的处理,消除回声信号 ^)和噪声 ,保留本端信号。 具体而言, 该方法包括以下步骤, 请参考图 4:
SI : AEC 模块对待处理音频信号进行初步回声 ·ί氏消处理, 生成初步回 声^^消信号
AEC模块对参考信号 x(«)、 待处理音频信号 《)进行处理, 输出初步回 声抵消信号 人造回声信号 («)以及频域的低带回声滤波系数 H( t) ; 并 将参考信号 χ(«)和待处理音频信号 《)输出到 RES模块。 具体而言, 包括:
1、 根据参考信号 x(«)生成人造回声信号 («)。
参考信号; )经过 QMF分析滤波器处理后生成高带参考信号 xhi (n)和低 带参考信号 w(«) , 其中, 输出高带参考信号 到 QMF合成滤波器, 输 出低带参考信号 w(«)到自适应滤波器;
低带参考信号 w(«)经过自适应滤波器处理后生成低带人造回声信号 yl'ow(n) , 并输出给 QMF合成滤波器;
高带参考信号 和低带人造回声信号 _y;。v )经过 QMF合成滤波器处 理后生成人造回声信号 («), 并输出给 RES模块;
2、 根据待处理音频信号 d(n)生成初步回声抵消信号 。
待处理音频信号 《)经过 QMF分析滤波器处理后, 生成高带待处理音 频信号 和低带待处理音频信号 其中, 输出高带待处理音频信号 du (n)到 QMF合成滤波
低带待处理音频信号 dlow («)减去低带人造回声信号 {n)得到低带初步 回声抵消信号 eto ) , 并输出到 QMF合成滤波模块;
高带待处理音频信号 O)和低带初步回声抵消信号 )经过 QMF合 成滤波器处理后, 生成初步回声抵消信号 并输出到 RES模块;
3、 频域的低带回声滤波系数 是由自适应滤波处理低带参考信号的 过程中计算得到的;
另夕卜, AEC模块还将参考信号 x(«)和待处理音频信号 输出到 RES模 块作为 RES模块经行进一步回声抵消的辅助处理信号。
S2: RES模块对初步回声抵消信号 e(n)进行进一步回声抵消处理, 生成 回声抵消后信号, 请参考图 5。
S201 : RES模块分别对参考信号 x(«)、 待处理音频信号 人造回声 信号 (")以及初步回声抵消信号 进行快速傅里叶变换( FFT , Fast Fourier Transformation ) , 生成对应的参考频域信号 Jr( t)、待处理频域音频信号 D k、、 人造回声频域信号 以及初步回声抵消频域信号 ; 并将待处理频域音 频信号 人造回声频域信号 :)、 以及残留频域回声信号 输入到 RES低带模块, 将参考频域信号 JT( t)、 低带回声滤波系数 H( t)以及初步回声 抵消频域信号 输入到 RES高带模块;
S203 : RES高带模块对初步回声抵消频域信号 £ )中的高带信号(以下 简称高带信号) 进行包络预测回声抑制, 计算并输出残留回声抑制系数 Ghe (k) ; RES 低带模块对初步回声抵消频域信号 中的低带信号 (以下简 称低带信号)釆用现有的回声抑制技术进行回声抑制, 输出处理结果; 请参 考图 6。
需要说明的是, 为了消除混叠影响, 高带信号与低带信号之间有重叠, 如果有 512根谱线, 低带 RES对 263根以下的谱线进行残留回声抑制, 而 高带对 231根以上的谱线进行抑制, 共有 32根谱线的重叠。
1、 RES高带模块计算残留回声抑制系数的过程为, 根据初步回声抵消 频域信号 的各子带能量 £( )、参考频域信号 JT( t)各子带能量 f/(0、残留回 声频域信号各子带能量 、 高带包络预测系数 gl (0以及高带包络预测系数 的衰减因子 factor 计算各子带增益; 根据各子带增益计算高带各谱线增益 Ghe (k) , 即残留回声系数。 具体而言, 包括:
( 1 )计算初步回声抵消频域信号 £ )的各子带能量 £( )、 参考频域信号
JT( t)各子带能量 f/(0、 残留回声频域信号各子带能量 (0、 高带包络预测系 数 gl (i)以及高带包络预测系数的衰减因子 factor。
A、高带 RES模块釆用包络预测的方法估计高带的回声,然后进行抑制。 所谓包络预测就是不考虑相位的影响,在频域上用前面若干帧参考信号的能 量来估计当前帧回声的能量, 用公式表示如下:
0 = 象 /(0 ( D
1=0
公式中, i表示在频域上划分的子带序号, L是用于包络预测的帧数。 ^() 表示第 i个子带第 1个系数, f/_,(0表示参考频域信号 前面第 1个帧第 i 个子带的能量。 在谱线数目为 512根时, 高带 RES的子带划分方式请参考 图 7, 共 5个子带, 子带之间釆用三角窗的方式进行重叠。
需要说明的是, 本例中子带之间釆用三角窗的方式进行重叠, 但不限於 此方式, 还可以釆用正弦窗、 矩形窗等方式进行重叠。
是第 i个子带估计出来的残留回声频域信号的能量。 如前文所述, 待处理音频信号 包括回声信号 本端信号 以及噪声 三部分。 待处理音频信号 《)经过 AEC模块初步回声抵消处理后, 生成了初步回声 抵消信号 该信号通过 FFT变换生成了初步回声抵消频域信号 所 以初步回声 4氏消频域信号 E<J 信号同样包括三部分: 残留回声频域信号、 本 端频域信号和噪声频域信号。
公式中 w«是第 i个子带的起始频率, (0第 i个子带的结束频率, 是各个频率的加权系数。 令 ^(0为第 i个子带的中心频率, 则^ ()按如下公 式取值:
(k-kl{i))
k l)一 k >= kl{i) & &k <= kc{i)
其中, i 取为 kc( - 1) , kh{i)取为 kc( + 1)。 这样取值可以增强声音的平 滑效果。
用上述计算 f/(0的方式可以计算初步回声抵消频域信号 £ )各个子带的
公式中 Ν是帧长。 由于只能获得频域的低带回声滤波系数 H(A), 因此 只能用低带的回声滤波系数来近似高带的滤波系数, 具体实现方法是, 将 HW进行 IFFT变换, 获得时域的低带冲激响应, 然后进行高通滤波, 获得 高子带的冲激响应 A 0„ , 然后用上面公式计算计算包络预测系数:
l*N+N-l
(6) n=l*N
从上面公式可知, 高带的各子带釆用相同的包络预测系数 t
D、 计算包络预测系数的衰减因子 factor。
图 8是回声冲激响应以及能量衰减曲线 (EDC)的一个示例:
(8)
= 10*logloiV
m=l 乂 对于帧 EDC, EDC斜率的含义为相邻帧 EDC的差值, 即:
q = EDC{l)-EDC{l-\) (9) q为 EDC斜率。 因此有:
相邻帧之间包络预测系数的衰减因子为: factor = gmlgm_l ( 11 ) 则有
= 10 * log 1 ^{factor)
因此通过计算出 EDC的斜率 q就可以计算出包络预测系数的衰减因子。
EDC 的斜率可以通过线性递归的方法计算。 可以看出, 除了起始的属于早 期回声的几十毫秒, EDC 是线性的, 即包络预测系数的衰减因子是一固定 值。用包络预测系数估计回声时用到的帧数比较多,例如回声拖尾为 256ms, 每帧的长度为 8ms, 则需要 32帧, 这样需要较多的存储空间和计算量, 而 利用衰减因子是一固定值这一特性, 可以减小存储空间和计算量, 具体方法 如下:
/0 L—1
1=0 1=10+1
ιο
( 13 )
公式中'。取为 7, 即前面 8帧 (64ms)认为是早期回声, 而 8帧之后的 回声认为是后期回声, 用 表示, m表示是当前帧, r表示是后期回声, 利用公式 factor = gl I gl_x , 可得出 ∞ ()的近似估计方式:
1-1
/=/0+1
1=10+2
= factor *
1
= factor * 1
其中, 为 ()的分贝值;
) = 10*logl0 ( 16) 为 £(0的分贝值。
G« = 0.1 , Edb(i)-Edb(i)<=0
G(i) = [Edb (0 - Edb {ι)Υ (1 - 0· 1)/10 + 0· 1 ,
Edb (0 - Edb (ι)〉 0) & Edb (ι) - Edb ( <10) ( 16) G{i) = l, Edb{i)-Edb{i)>=\0
上面公式中, 表示各子带增益, 6(0<=ο表示估计的回声比 原始的回声还要大, 此时子带增益取下限 0.1; Jd6b(,o>=io表示原始回
声比估计回声大 10DB以上, 此时可以基本认为麦克风信号里面没有回声, 子带增益取上限 1; (ΕΛ(ή - db(p> > w & (ΕΛ(ή - db(p> < 表示原始回声比估计 回声大 0DB至 10DB之间, 此时子带增益取值范围为 0.1至 1之间, 且和
{Edb(j)-Edb(i))的 DB差值成正比。
(3)在计算出子带增益之后, 高带各谱线的增益 即残留回声抑 制系数可用相邻子带增益插值获得:
G(k) = wk(i)G(i) + wk(i-l)G(i-l) (17) wk ()就是前面计算各子带能量的系数。
S205: RES低带模块输出结果乘以 RES高带模块输出的残留回声抑制 系数, 得到并输出回声抵消后信号 £»。
S207:对回声抵消后信号 £' (A)进行快速傅里叶反变换( IFFT, Inverse Fast Fourier Transformation ) , 再回填噪声, 输出最终处理信号。
需要说明的是, 其中回填噪声模块根据 NLP标识判断当前帧是是否是 残留回声, 如果是, 则回填噪声, 否则判断 VAD标识, 如果当前帧是噪声, 则保存噪声信号到噪声緩冲区。
本发明实施例通过将信号划分为高带信号和低带信号,对低带信号釆用 现有的回声抵消算法, 对高带信号釆用包络预测的回声抑制算法, 达到了大 幅降低计算复杂度的效果, 同时可以避免子带自适应滤波边界泄露的问题。
本发明实施例同时提供一种回声抵消 AEC模块, 包括
参考信号分析滤波器, 将参考信号处理为高带参考信号和低带参考信 号, 将高带参考信号输出到人造回声合成滤波器, 将低带参考信号输出到自 适应滤波器;
待处理音频信号分析滤波器,将待处理音频信号处理为高带待处理音频 信号和低带待处理音频信号,将高带待处理音频信号输出到初步回声抵消信 号合成滤波器; 将低带待处理音频信号输出到自适应滤波器;
自适应滤波器,与参考信号分析滤波器和待处理音频信号分析滤波器相
连, 将低带参考信号经过自适应滤波处理生成低带人造回声信号, 并输出到 人造回声合成滤波器;将低带待处理音频信号经过自适应滤波处理生成低带 初步回声抵消信号, 并输出到初步回声抵消信号合成滤波器;
人造回声合成滤波器, 与自适应滤波器相连, 将高带参考信号和低带人 造回声信号合成为人造回声信号, 并输出给 RES设备;
初步回声抵消信号合成滤波器, 与自适应滤波器相连, 将高带待处理音 频信号和低带初步回声抵消信号合成为初步回声抵消信号, 并输出给 RES 模块。
需要说明的是, QMF 的分析和综合滤波器处理流程如图 9 , 输入信号 x(n)被低通滤波器 H0(z)和高通滤波器 Hl(z)滤波之后, 进行 2倍下釆样, 然 后分别进行低带处理和高带处理, 再分别进行 2倍上釆样, 以及综合滤波器 F0(z)和 Fl(z)滤波, 相加之后输出合成的信号 y(n)。 为了消除混叠, 滤波器 有以下一些限制:
Hl(z) = H0(-z)
G0(z) = 2*Η1(-ζ) = 2*H0(z)
Gl(z) = -2*H0(-z) = -2*Hl(z)
在时域上, 令 hO(i) = h(i), i= 0,l ...N-l。 其中 N为滤波器的长度, 则有 hl(i) = (-^ hCi)
gO(i) = 2h(i)
gl(i) = -2* (-1)! h(i)
因此, 在设计滤波器时只需要设计低通分析滤波器就行了, 其他滤波器 可以根据低通分析滤波器参考上述参数来获得。
本发明实施例同时还提供一种回声抵消 RES模块, 包括:
FFT变换模块, 将参考信号、 待处理音频信号、 人造回声信号、 初步回 声抵消信号经过 FFT 变换, 生成对应的参考频域信号、 待处理频域音频信 号、 人造回声频域信号以及初步回声抵消频域信号; 并将待处理频域音频信
号、 人造回声频域信号以及初步回声抵消频域信号输入到 RES低带模块, 将参考频域信号、 初步回声抵消频域信号输入到 RES高带模块;
RE S高带模块,对初步回声抵消频域信号中的高带信号进行包络预测回 声抑制, 计算并输出残留回声抑制系数;
RES低带模块, 对初步回声抵消频域信号中的低带信号进行回声抑制, 输出处理结果。
其中, RES高带模块包括:
初步回声抵消频域信号各子带能量计算模块,计算初步回声抵消频域信 号各子带能量,并输出所述初步回声抵消频域信号子带能量到子带增益计算 模块;
参考频域信号各子带能量计算模块, 用于计算参考频域信号各子带能 量, 并输出所述参考频域信号各子带能量到所述子带增益计算模块;
IFFT变换模块, 将低带回声滤波系数进行 IFFT变换, 生成时域的低带 冲激响应, 并将所述冲激响应输出到高通滤波模块;
所述高通滤波模块, 与所述 IFFT变换模块相连, 将所述时域的低带冲 激响应进行高通滤波处理, 生成高带的冲激响应, 并将所述高带的冲激响应 输出到高带包络预测系数计算模块;
所述高带包络预测系数计算模块, 与所述高通滤波模块相连, 根据所述 高带的冲激响应计算高带包络预测系数,并将所述高带包络预测系数输出到 EDC斜率计算模块和所述子带增益计算模块;
所述 EDC斜率计算模块, 与所述高带包络预测系数计算模块相连, 计 算 EDC斜率 , 并将所述 EDC斜率输出到衰减因子计算模块;
所述衰减因子计算模块,与所述 EDC斜率计算模块相连,根据所述 EDC 斜率计算衰减因子, 并将所述衰减因子输出到所述子带增益计算模块;
所述子带增益计算模块,与所述初步回声抵消频域信号各子带能量计算 模块、 参考频域信号各子带能量计算模块、 高带包络预测系数计算模块以及 衰减因子计算模块相连, 根据所述初步回声抵消频域信号各子带能量、 参考
频域信号各子带能量、 高带包络预测系数以及衰减因子计算子带增益, 并将 所述子带增益输出到高带各谱线增益计算模块;
所述高带各谱线增益计算模块, 与所诉子带增益计算模块相连, 根据所 述子带增益计算高带各谱线增益, 即残留回声抑制系数。
本发明实施例提供的设备通过将信号划分为高带信号和低带信号,对低 带信号釆用现有的回声抵消算法, 对高带信号釆用包络预测的回声抑制算 法, 达到了大幅降低计算复杂度的效果, 同时可以避免子带自适应滤波边界 泄露的问题。
最后应说明的是: 以上实施例仅用以说明本发明的技术方案, 而非对其 限制; 尽管参照前述实施例对本发明进行了详细的说明, 本领域的普通技术 人员应当理解: 其依然可以对前述各实施例所记载的技术方案进行修改, 或 者对其中部分技术特征进行等同替换; 而这些修改或者替换, 并不使相应技 术方案的本质脱离本发明各实施例技术方案的精神和范围。
Claims
1、 一种回声 4氏消方法, 其特征在于, 所述方法包括以下步骤: 划分待处理音频信号为高带待处理音频信号和低带待处理音频信号; 对所述低带待处理音频信号进行自适应滤波回声抵消处理, 对所述高带待 处理音频信号不进行自适应滤波回声抵消处理, 生成初步回声抵消信号; 对初步回声抵消频域信号中的高带信号进行包络预测回声抑制, 计算并输 出残留回声抑制系数; RES模块对初步回声抵消频域信号中的低带信号进行回 声抑制, 输出处理结果; 将所述输出结果与残留回声抑制系数相乘输出回声抵 消后信号。
2、 根据权利要求 1所述的回声抵消方法, 其特征在于, 所述划分待处理音 频信号是通过 QMF分析滤波进行划分。
3、 根据权利要求 1所述的回声抵消方法, 其特征在于, 所述步骤 B包括: 低带待处理音频信号经过自适应滤波处理后生成低带初步回声抵消信号; 所述低带初步回声抵消信号和高带待处理音频信号进行合成滤波处理之后 生成初步回声抵消信号;
4、 根据权利要求 1或 3所述的回声抵消方法, 其特征在于, 所述步骤 B 还包括:
AEC模块将参考音频信号通过 QMF分析滤波, 生成高带参考音频信号和 低带参考音频信号;
AEC模块将所述低带参考音频信号进行自适应滤波处理生成低带合成回声 信号;
AEC模块将所述高带参考音频信号和所述低带合成回声信号进行 QMF合 成滤波处理, 生成合成回声信号;
5、 根据权利要求 1所述的回声抵消方法, 其特征在于, 进行所述所述步骤 C之前还包括: 将所述初步回声 ·ί氏消信号经过 FFT变换生成初步回声 ·ί氏消频域信号。
6、 根据权利要求 1所述的回声抵消方法, 其特征在于, 所述计算残留回声 抑制系数包括: 计算初步回声抵消频域信号各子带能量, 计算参考频域信号各子带能量, 计算残留回声频域信号各子带能量, 计算高带包络预测系数, 计算高带包络预 测系数的衰减因子; 根据所述初步回声抵消频域信号各子带能量、 参考频域信号各子带能量、 残留回声频域信号各子带能量、 初步回声抵消频域信号中的高带信号的包络预 测系数以及高带包络预测系数的衰减因子计算初步回声抵消频域信号中的高带 信号的各子带增益; 根据初步回声抵消频域信号中的高带信号的各子带增益计算初步回声抵消 频域信号中的高带信号的各语线增益, 即残留回声抑制系数。
7、 根据权利要求 6所述的回声抵消方法, 其特征在于, 所述计算初步回声 抵消频域信号中的高带信号的包络预测系数, 方法为: l*N+N—\
根据公式 g ')= ∑ ') * ')计算所述包络预测系数;
n=l*N 其中& (0表示所述包络预测系数, Ν表示帧长, /7„( )表示初步回声抵消频 域信号中的高带信号各子带的回声冲激响应;
获得 θ的方法为,将低带回声滤波系数 HW进行 IFFT变换,获得时域的 低带冲激响应, 然后将所述低带冲激响应进行高通滤波处理, 获得高带的冲激 响 hn (i)。
8、 根据权利要求 6所述的回声抵消方法, 其特征在于, 所述计算各子带能 量方法为: 计算早期回声子带能量, 计算后期回声子带能量; 将早期回声子带能量与后期回声子带能量相加得到子带能量。
9、 根据权利要求 1所述的回声抵消方法, 其特征在于, 所述初步回声抵消 频域信号中的高带信号与初步回声抵消频域信号中的低带信号,二者的关系为, 二者之间有重叠, 以消除混叠影响。
10、 根据权利要求 1、 2和 3项中任一所述的回声抵消方法, 其特征在于, 所述将待处理音频信号划分为高带待处理音频信号和低带待处理音频信号之 前, 使用音频釆集设备釆集所述待处理音频信号, 所述待处理音频信号包括: 回声音频信号、 本端信号以及噪声信号。
11、 一种回声抵消 AEC模块, 其特征在于, 所述 AEC模块包括: 参考信号分析滤波器, 将参考信号处理为高带参考信号和低带参考信号, 将高带参考信号输出到人造回声合成滤波器, 将低带参考信号输出到自适应滤 波器; 待处理音频信号分析滤波器, 将待处理音频信号处理为高带待处理音频信 号和低带待处理音频信号, 将高带待处理音频信号输出到初步回声抵消信号合 成滤波器; 将低带待处理音频信号输出到自适应滤波器; 所述自适应滤波器, 与所述参考信号分析滤波器和待处理音频信号分析滤
波器相连, 将所述低带参考信号经过自适应滤波处理生成低带人造回声信号, 并输出到人造回声合成滤波器; 将所述低带待处理音频信号经过自适应滤波处 理生成低带初步回声抵消信号, 并输出到初步回声抵消信号合成滤波器; 所述人造回声合成滤波器, 与所述自适应滤波器相连, 将所述高带参考信 号和低带人造回声信号合成为人造回声信号, 并输出给 RES设备; 所述初步回声抵消信号合成滤波器, 与所述自适应滤波器相连, 将所述高 带待处理音频信号和低带初步回声抵消信号合成为初步回声抵消信号, 并输出 给 RES模块。
12、 一种回声抵消 RES模块, 其特征在于, 所述 RES模块包括:
FFT 变换模块, 将参考信号、 待处理音频信号、 人造回声信号、 初步回声 抵消信号经过 FFT变换, 生成对应的参考频域信号、 待处理频域音频信号、 人 造回声频域信号以及初步回声 ·ί氏消频域信号; 并将待处理频域音频信号、 人造 回声频域信号以及初步回声抵消频域信号输入到 RES低带模块,将参考频域信 号、 初步回声抵消频域信号输入到 RES高带模块;
RES高带模块, 对初步回声抵消频域信号中的高带信号进行包络预测回声 抑制, 计算并输出残留回声抑制系数;
RES低带模块, 对初步回声抵消频域信号中的低带信号进行回声抑制, 输 出处理结果。
13、 根据权利要求 12所述的 RES模块, 其特征在于, 所述 RES高带模块 包括: 初步回声抵消频域信号各子带能量计算模块, 计算初步回声抵消频域信号 各子带能量,并输出所述初步回声抵消频域信号子带能量到子带增益计算模块; 参考频域信号各子带能量计算模块, 用于计算参考频域信号各子带能量,
并输出所述参考频域信号各子带能量到所述子带增益计算模块;
IFFT变换模块, 将低带回声滤波系数进行 IFFT变换, 生成时域的低带冲 激响应, 并将所述冲激响应输出到高通滤波模块; 所述高通滤波模块, 与所述 IFFT变换模块相连, 将所述时域的低带冲激响 应进行高通滤波处理, 生成高带的冲激响应, 并将所述高带的冲激响应输出到 高带包络预测系数计算模块; 所述高带包络预测系数计算模块, 与所述高通滤波模块相连, 根据所述高 带的冲激响应计算高带包络预测系数, 并将所述高带包络预测系数输出到 EDC 斜率计算模块和所述子带增益计算模块; 所述 EDC斜率计算模块, 与所述高带包络预测系数计算模块相连, 计算 EDC斜率, 并将所述 EDC斜率输出到衰减因子计算模块; 所述衰减因子计算模块, 与所述 EDC斜率计算模块相连, 根据所述 EDC 斜率计算衰减因子, 并将所述衰减因子输出到所述子带增益计算模块; 所述子带增益计算模块, 与所述初步回声抵消频域信号各子带能量计算模 块、 参考频域信号各子带能量计算模块、 高带包络预测系数计算模块以及衰减 因子计算模块相连, 根据所述初步回声抵消频域信号各子带能量、 参考频域信 号各子带能量、 高带包络预测系数以及衰减因子计算子带增益, 并将所述子带 增益输出到高带各语线增益计算模块; 所述高带各谱线增益计算模块, 与所诉子带增益计算模块相连, 根据所述 子带增益计算高带各语线增益, 即残留回声抑制系数。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| EP13845843.5A EP2905778B1 (en) | 2012-10-12 | 2013-06-04 | Echo cancellation method and device |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210387313.8A CN103730125B (zh) | 2012-10-12 | 2012-10-12 | 一种回声抵消方法和设备 |
| CN201210387313.8 | 2012-10-12 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2014056328A1 true WO2014056328A1 (zh) | 2014-04-17 |
Family
ID=50454171
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2013/076685 Ceased WO2014056328A1 (zh) | 2012-10-12 | 2013-06-04 | 一种回声抵消方法和设备 |
Country Status (4)
| Country | Link |
|---|---|
| US (1) | US8811624B2 (zh) |
| EP (1) | EP2905778B1 (zh) |
| CN (1) | CN103730125B (zh) |
| WO (1) | WO2014056328A1 (zh) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10504538B2 (en) | 2017-06-01 | 2019-12-10 | Sorenson Ip Holdings, Llc | Noise reduction by application of two thresholds in each frequency band in audio signals |
Families Citing this family (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9319783B1 (en) * | 2014-02-19 | 2016-04-19 | Amazon Technologies, Inc. | Attenuation of output audio based on residual echo |
| US9363600B2 (en) | 2014-05-28 | 2016-06-07 | Apple Inc. | Method and apparatus for improved residual echo suppression and flexible tradeoffs in near-end distortion and echo reduction |
| FR3025923A1 (fr) * | 2014-09-12 | 2016-03-18 | Orange | Discrimination et attenuation de pre-echos dans un signal audionumerique |
| GB2545263B (en) * | 2015-12-11 | 2019-05-15 | Acano Uk Ltd | Joint acoustic echo control and adaptive array processing |
| US10122863B2 (en) * | 2016-09-13 | 2018-11-06 | Microsemi Semiconductor (U.S.) Inc. | Full duplex voice communication system and method |
| CN106601266A (zh) * | 2017-02-24 | 2017-04-26 | 喜大(上海)网络科技有限公司 | 回声消除方法、装置及系统 |
| CN109308905B (zh) * | 2017-07-28 | 2021-08-20 | 北京搜狗科技发展有限公司 | 音频数据处理方法、装置、电子设备和存储介质 |
| CN109524018B (zh) * | 2017-09-19 | 2022-06-10 | 华为技术有限公司 | 一种回声处理方法及设备 |
| CN109658946A (zh) * | 2017-10-12 | 2019-04-19 | 深圳前海黑鲸科技有限公司 | 一种回声处理方法、装置、存储介质及终端设备 |
| CN110176244B (zh) * | 2018-06-19 | 2023-10-03 | 腾讯科技(深圳)有限公司 | 回声消除方法、装置、存储介质和计算机设备 |
| CN108962276B (zh) * | 2018-07-24 | 2020-11-17 | 杭州听测科技有限公司 | 一种语音分离方法及装置 |
| US10867617B1 (en) * | 2018-12-10 | 2020-12-15 | Amazon Technologies, Inc. | Techniques for processing audio data |
| CN110286813B (zh) | 2019-05-22 | 2020-12-01 | 北京达佳互联信息技术有限公司 | 图标位置确定方法和装置 |
| CN110246516B (zh) * | 2019-07-25 | 2022-06-17 | 福建师范大学福清分校 | 一种语音通信中小空间回声信号的处理方法 |
| CN112491443B (zh) * | 2019-09-11 | 2022-10-28 | 瑞昱半导体股份有限公司 | 具有回音消除省电机制的电子系统及相关回音消除方法 |
| CN110719546A (zh) * | 2019-09-25 | 2020-01-21 | 成都九洲迪飞科技有限责任公司 | 一种嵌入式数字会议方法以及会议系统 |
| CN111028855B (zh) * | 2019-12-17 | 2022-07-08 | 广州视源电子科技股份有限公司 | 回声抑制方法、装置、设备及存储介质 |
| CN113194385B (zh) * | 2021-01-14 | 2023-03-10 | 四川湖山电器股份有限公司 | 基于步长控制的子带自适应反馈消除方法及系统 |
| CN113019479A (zh) * | 2021-03-31 | 2021-06-25 | 中国人民解放军空军军医大学 | 一种用于模拟井下工作环境的试验箱 |
| CN115359788A (zh) * | 2022-07-20 | 2022-11-18 | 海信视像科技股份有限公司 | 一种显示装置和远场语音识别方法 |
| CN115527550B (zh) * | 2022-08-23 | 2026-01-16 | 全时云商务服务股份有限公司 | 一种单麦克风子带域降噪方法及系统 |
| CN121725793A (zh) * | 2026-02-25 | 2026-03-24 | 上海声网科技有限公司 | 基于声纹分析的智慧家居声控用户识别方法及系统 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003284184A (ja) * | 2002-03-25 | 2003-10-03 | Sony Corp | エコーキャンセラ及びエコーキャンセリング方法 |
| CN1805011A (zh) * | 2005-12-23 | 2006-07-19 | 北京中星微电子有限公司 | 一种提高移动通信设备语音质量的自适应滤波方法及装置 |
| WO2009151062A1 (ja) * | 2008-06-10 | 2009-12-17 | ヤマハ株式会社 | 音響エコーキャンセラおよび音響エコーキャンセル方法 |
| CN101778183A (zh) * | 2009-01-13 | 2010-07-14 | 华为终端有限公司 | 一种残留回声抑制方法及设备 |
| CN102379004A (zh) * | 2009-04-03 | 2012-03-14 | 株式会社Ntt都科摩 | 语音编码装置、语音解码装置、语音编码方法、语音解码方法、语音编码程序以及语音解码程序 |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100842590B1 (ko) * | 2004-11-09 | 2008-07-01 | 삼성전자주식회사 | 이동 단말에서 음향 반향 제거 방법 및 장치 |
| US7783478B2 (en) | 2007-01-03 | 2010-08-24 | Alexander Goldin | Two stage frequency subband decomposition |
| DE102008039329A1 (de) * | 2008-01-25 | 2009-07-30 | Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. | Vorrichtung und Verfahren zur Berechnung von Steuerinformationen für ein Echounterdrückungsfilter und Vorrichtung und Verfahren zur Berechnung eines Verzögerungswerts |
| JP5424936B2 (ja) * | 2010-02-24 | 2014-02-26 | パナソニック株式会社 | 通信端末及び通信方法 |
-
2012
- 2012-10-12 CN CN201210387313.8A patent/CN103730125B/zh active Active
-
2013
- 2013-06-04 WO PCT/CN2013/076685 patent/WO2014056328A1/zh not_active Ceased
- 2013-06-04 EP EP13845843.5A patent/EP2905778B1/en active Active
- 2013-08-30 US US14/015,319 patent/US8811624B2/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003284184A (ja) * | 2002-03-25 | 2003-10-03 | Sony Corp | エコーキャンセラ及びエコーキャンセリング方法 |
| CN1805011A (zh) * | 2005-12-23 | 2006-07-19 | 北京中星微电子有限公司 | 一种提高移动通信设备语音质量的自适应滤波方法及装置 |
| WO2009151062A1 (ja) * | 2008-06-10 | 2009-12-17 | ヤマハ株式会社 | 音響エコーキャンセラおよび音響エコーキャンセル方法 |
| CN101778183A (zh) * | 2009-01-13 | 2010-07-14 | 华为终端有限公司 | 一种残留回声抑制方法及设备 |
| CN102379004A (zh) * | 2009-04-03 | 2012-03-14 | 株式会社Ntt都科摩 | 语音编码装置、语音解码装置、语音编码方法、语音解码方法、语音编码程序以及语音解码程序 |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10504538B2 (en) | 2017-06-01 | 2019-12-10 | Sorenson Ip Holdings, Llc | Noise reduction by application of two thresholds in each frequency band in audio signals |
Also Published As
| Publication number | Publication date |
|---|---|
| EP2905778A1 (en) | 2015-08-12 |
| US8811624B2 (en) | 2014-08-19 |
| EP2905778B1 (en) | 2018-05-09 |
| EP2905778A4 (en) | 2015-11-25 |
| US20140105410A1 (en) | 2014-04-17 |
| CN103730125A (zh) | 2014-04-16 |
| CN103730125B (zh) | 2016-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2014056328A1 (zh) | 一种回声抵消方法和设备 | |
| JP5671147B2 (ja) | 後期残響成分のモデリングを含むエコー抑制 | |
| US9992572B2 (en) | Dereverberation system for use in a signal processing apparatus | |
| JP6138910B2 (ja) | アクティブノイズキャンセル適応フィルタのためのプレシェーピング直列フィルタ | |
| CN102347028A (zh) | 双麦克风语音增强装置及方法 | |
| US8751224B2 (en) | Combined microphone and earphone audio headset having means for denoising a near speech signal, in particular for a “hands-free” telephony system | |
| AU756511B2 (en) | Signal noise reduction by spectral subtraction using linear convolution and causal filtering | |
| CN1175709C (zh) | 利用频谱减除的双拾音器信号降噪系统和方法 | |
| CN102074245B (zh) | 基于双麦克风语音增强装置及语音增强方法 | |
| CN108376548B (zh) | 一种基于麦克风阵列的回声消除方法与系统 | |
| CN110956975B (zh) | 回声消除方法及装置 | |
| CN101976565A (zh) | 基于双麦克风语音增强装置及方法 | |
| CN102347027A (zh) | 双麦克风语音增强装置及其语音增强方法 | |
| CN102739886A (zh) | 基于回声频谱估计和语音存在概率的立体声回声抵消方法 | |
| CN103428385A (zh) | 用于处理音频信号的方法及用于处理音频信号的电路布置 | |
| WO2012142270A1 (en) | Systems, methods, apparatus, and computer readable media for equalization | |
| WO2011094232A1 (en) | Adaptive noise reduction using level cues | |
| CN101933088A (zh) | 用于回声抑制的滤波器系数的计算设备和方法 | |
| JP2007129736A (ja) | 音声信号中の暗騒音を抑制する方法及び装置、並びに反響消去を伴う対応する装置 | |
| CN102074246A (zh) | 基于双麦克风语音增强装置及方法 | |
| US9390718B2 (en) | Audio signal restoration device and audio signal restoration method | |
| EP1080463A1 (en) | Signal noise reduction by spectral subtraction using spectrum dependent exponential gain function averaging | |
| CN106448691A (zh) | 一种用于扩音通信系统的语音增强方法 | |
| JP2006067127A (ja) | 残響除去方法及びその装置 | |
| CN111917926A (zh) | 一种通信终端中的回声消除方法、装置及终端设备 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 13845843 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2013845843 Country of ref document: EP |






