JP2003533902A - Controlling echo in the encoded domain - Google Patents

Abstract

(57) Abstract: A communication system (10) transmits a close-terminated digital signal using a compression code having a plurality of parameters including a first parameter. This parameter is representative of a single audio signal having multiple audio features. The system also transmits the digital signal at the remote end using a code of compression. Terminal (20) receives the near-terminated digital signal and terminal (36) receives the remote-terminated digital signal. The processor (40) reads at least the first parameter in response to the digital signal of the proximity termination. A processor generates at least partially decoded near-terminated signals and at least partially decoded remote-terminated signals. Based on such a signal, the processor adjusts the first parameter and writes the adjusted first parameter to the near-terminated digital signal. The other terminal (22) transmits the adjusted near-terminated digital signal. As a result, echoes in the digital signal at the near end are reduced.

Description

Detailed Description of the Invention

Description of Related Application This is a utility application corresponding to Provisional Application No. 60 / 142,136, filed July 2, 1990, entitled "Coding Domain Enhancement of Compressed Speech."

[0002] Federal Government Research and Development Declaration   None applicable

BACKGROUND OF THE INVENTION The present invention relates to coded domain enhancement of compressed speech, and more particularly to coded domain echo control.

This specification refers to the following references. [1] GSM 06.10 "Digital Cellular Communication System (Phase 2); Full Rate Speech; Part 2: Trans Coding", ETS 300 580-2, March 1998, Second Edition. [2] GSM 06.60 "Digital Cellular Communication System (Phase 2); Enhanced Full Rate (EFR) Voice Trans Coding", June 1998. [3] GSM 08.62 “Digital Cellular Communication System (Phase 2+); In-Band Tandem Free Operation (TFO) of Voice Encoder”, ETSI, 2000
March. [4] JRDeller, JGProakis, JHLHansen "Individual time processing of audio signals",
Chapter 7, Prentice-Hall Inc., 1987. [5] GSM 06.12 "European Digital Cellular Communication System (Phase 2); Suitable Noise Surface for Full Rate Voice Traffic Channels", ETSI, 1994.

In GSM digital cellular networks, voice transmission between a mobile station (handset) and a base station takes place in compressed or coded form. GSM FR [1
] Or EFR [2] or other audio coding technique is used to compress the audio. The device used for voice compression is called a vocoder. The number of bits required for coded speech is less than 2 bits per sample. This situation is depicted in Figure 1. Voice is transmitted in uncoded form between base stations (using PCM companding, which requires 8 bits per sample).

The terms coded speech and non-coded speech can be explained as follows. Uncoded Speech: Refers to digital speech signal samples typically used in telephones. These samples are either linear 13 bits per sample or 8 bits per sample in a compounded form such as μ-law or A-law PCM, with a typical bit rate of 64 kbps.

Coded voice: 13 kbps for GSM FR, 12.2 kb for GSM EFR
Refers to compressed on-signal signal parameters (also referred to as encoding parameters), which typically use bit rates much lower than 64 kbps, such as ps. The compression method is more expensive than the simple PCM companding scheme. By way of example, compression methods are linear predictive coding, code-excited linear prediction and multiband excitation coding [4].
].

The Tandem Free Operations (TFO) standard [3] will be in the near future in GSM
Deployed in digital cellular networks. The TFO standard applies to mobile cross-calls. Under TFO, audio signals are transported between mobiles in compressed form after a short negotiation period. This eliminates the tandem voice code during mobile inter-calls. The elimination of tandem codes is known to improve voice quality when the original signal is clear. The key point to note is that the voice transmission remains encoded between the mobile handsets, as depicted in FIG.

Under TFO, the transmission between the handset and the base station is coded, requiring less than 2 bits per voice sample. However, 8 bits per voice sample are still available for transmission between base stations. At the base station, 8 bits per sample are needed because the speech is decoded and then A-law expanded. Nevertheless, the original coded speech bits are each A-law compounded 8
Used to replace the two least significant bits (LSBs) in a bit sample. Once the TFO is established between the handsets, the base station only sends two LSBs in each 8-bit sample to each handset, discarding the six most significant bits. This avoids tandeming the vocoder. This process is depicted in Figure 3.

The echo problem and its traditional solution are shown in FIG. In a wired network,
Echoes are generated by impedance mismatch in a 4-wire vs. 2-wire hybrid. The mismatch results in a portion of the far-end signal being electrically reflected in the near-end signal. Depending on the network delay and the channel impulse response of the end path, the echo can be offensive to the far end listener. The end-path impulse response is evaluated by the echo canceller (EC) of the network and used to make the echo signal estimate. The evaluation result is then subtracted from the near-end signal to remove the echo. After EC processing, any residual echo is removed by a non-linear processor (NLP).

In the case of digital cellular handsets, echoes are generated by feedback from a speaker (speaker) to a microphone (earpiece). Acoustic feedback can be significant, and echoes can be jarring, especially for hands-free phones.

FIG. 5 shows the feedback path from the speaker to the microphone in a digital cellular handset. The illustrated handset does not have the error cancellation function implemented in the handset.

Under TFO in a GSM network, if the echo cancellation function is to be implemented in the network, decoding of the coded speech, processing of the resulting uncoded speech, and its Re-encoding is required. Such decoding and re-encoding is necessary because traditional echo cancellers only work on uncoded speech signals. This approach is shown in FIG. Some of the drawbacks of this approach are:

1. This approach requires two decoders and one encoder, which is a significant amount of computation. Typically, encoders are one order of magnitude more computationally complex than decoders. Therefore, the existence of the encoder puts a heavy burden on the computer. 2. The delay introduced by the decoding and recoding process is undesirable. 3. A vocoder tandem (ie two pairs of encoders / decoders placed in series) was introduced into this approach, which, as is well known, would result in poor speech quality due to quantization effects.

In another straightforward approach, comfort noise generation may be used to mask the echo. Comfort noise generation is used for silence suppression or intermittent transmission purposes (eg [5]). Using such techniques, it is possible to completely mask the echo as soon as it is detected. However, such techniques have the drawback of "chopping", especially during double-talk conditions, as well as poor transparency and unnaturalness of the background.

The proposed technique can perform echo control directly on coded speech (ie by direct modification of coding parameters). Minimal computational complexity and delay. Tandemization effects are avoided or minimized, resulting in a better perceived sound quality after echo control. Also, excellent background transparency is achieved.

Speech compression belongs to the category of lossy source coding and is commonly referred to as speech coding. Speech coding is done to minimize the bandwidth required for speech transmission. This is especially important in radio telephones where bandwidth is scarce. In relatively bandwidth-rich packet networks, voice coding is still important in minimizing network delay and jitter. This is because voice communication, unlike data communication, can hardly tolerate delay. Therefore, the smaller the packet size, the easier the transmission via the packet network. The four relevant ETSI GSM standards are listed in Table 1.

Table 1: GSM Voice Codec Codec Name Coding Method Bit Rate (kbits / sec) Half Rate (HR) VSELP 5.6 Full Rate (FR) RPE-LTP 13 Enhanced Full Rate (EFR) ACELP 12.2 Adaptive Multi-Rate (AMR) MR-ACELP 5.4-12.2.

In speech coding, a set of consecutive digital speech samples is called a speech frame. The GSM encoder works on a frame size of 20 ms (160 samples at a sampling rate of 8 kHz). Given a single speech frame, the speech coder determines a small set of parameters for the speech synthesis model. With this voice parameter and voice synthesis model, it is possible to reconstruct a voice frame that appears in a form very similar to the original voice frame and emits a sound that is very similar. This reconstruction is done by the speech decoder. In the GSM vocoders listed above, the encoding process is much more computationally intensive than the decoding process.

The speech parameters determined by the speech coder depend on the speech synthesis model used. The GSM encoders listed in Table 1 are linear predictive coding (LPC).
Use the model. A simplified block diagram of the generic LPC speech synthesis model is shown in FIG. This model can be used to generate a voice-like signal by specifying model parameters accordingly. In the speech synthesis model of this example,
The parameters include time-varying filter coefficients, pitch period, codebook vector and gain coefficient. Synthetic speech is generated as follows. The codebook vector c (n) is first set to an appropriate size by the codebook gain coefficient G.
Here, n represents the sample time.

The determined codebook vector is then filtered by a pitch synthesis filter. The parameters of this filter include pitch gain g and pitch period T. The filtration result is sometimes referred to as the total excitation vector u (n). As the name implies, pitch synthesis filters provide the harmonic quality of emitted speech. The total excitation vector is then filtered by an LPC synthesis filter that specifies the broad spectrum shape of the speech frame and the corresponding broad spectrum shape of the audio signal.

The parameters are usually updated twice or more for each voice frame. For example GSM
In the FR encoder and the EFR encoder, the codebook vector, codebook gain, and pitch synthesis filter parameter are determined for each subframe (5 ms). The LPC synthesis filter parameter is set to twice (10 ms) per frame in EFR.
Each time), FR determines once per frame.

A typical sequence of steps used in a speech coder is as follows. 1. Get a frame of audio samples. 2. Multiply the window (eg Hamming window) on the sample frame,
Determine the autocorrelation function up to lag M. 3. The reflection coefficient and / or LPC coefficient is determined from the autocorrelation function. (Note that reflection coefficient is another way of saying LPC filter coefficients.)

[0024] 4. The reflection coefficient, or LPC filter coefficient, is transformed into another form suitable for quantization (eg log area ratio or line spectral frequency). 5. The transformed LPC coefficient is quantized using a vector quantization technique.

6. Add some auxiliary error correction / detection bits, framing bits, etc. 7. Transmit the encoded parameters.

The following operation sequence is typically performed by the speech coder for each subframe. 1. Determine the pitch period. 2. Determine the corresponding pitch gain. 3. Quantize pitch period and pitch gain.

[0027] 4. Back-filter the original audio signal through a quantized LPC synthesis filter
Acquire the residual signal. 5. The LPC residual signal is back-filtered through the pitch synthesis filter to obtain the pitch residual. 6. Determine the best codebook vector.

[0028] 7. Determine the best codebook gain. 8. Quantize codebook gains and codebook vectors. 9. Update the filter memory as appropriate.

A typical sequence of steps used in a speech decoder is as follows. First, some error correction / detection and framing is performed. Next, the following is executed for each subframe.

1. Dequantize all received coding parameters (LPC coefficients, pitch period, pitch gain, codebook vector, codebook gain). 2. The size of the codebook vector is defined by the codebook gain, which is filtered with a pitch synthesis filter to obtain the LPC excitation signal. 3. The LPC excitation signal is filtered by an LPC synthesis filter to obtain a preliminary voice signal.

4. Build a post filter (typically based on LPC coefficients). 5. Quantization noise is reduced by filtering the preliminary speech signal, thus obtaining the final synthesized speech.

As an example of the placement of the coding parameters in the bitstream transmitted by the encoder, consider the GSM FR vocoder. For the GSM FR vocoder, a frame is defined as 160 voice samples sampled at 8 kHz. That is, the frame is 20 ms long. With A-law PCM companding, this would require 1280 bits for the transmission of 160 samples. The encoder compresses 160 samples to 260 bits. The placement of various coding parameters within the 260 bits of each frame is shown in FIG.

The first 36 bits of each encoded frame consist of a log area ratio corresponding to the LPC synthesis filter. The remaining 224 bits can be classified into 4 subframes of 56 bits each. Within each subframe, the coding parameter bits first include pitch synthesis filter related parameters, followed by codebook vector related parameters and codebook gain related parameters.

SUMMARY OF THE INVENTION The preferred embodiment is useful in a communication system for transmitting near-end digital signals using a compressed code consisting of a plurality of parameters including a first parameter. The parameter represents an audio signal having a plurality of audio characteristics. The compressed code can be decoded in multiple decoding steps. The communication system also transmits the far end digital signal using the compressed code. In such an environment,
The echo in the near-end digital signal is at least the first of the plurality of parameters.
The parameter can be reduced by reading in response to the near-end digital signal. At least one of the plurality of decoding steps is performed on the near-end digital signal and the far-end digital signal to produce an at least partially decoded near-end signal and an at least partially decoded far-end signal. Will be.

The first parameter is adjusted in response to the at least partially decoded near-end signal and the at least partially decoded far-end signal, thereby producing an adjusted first parameter. Become. The first parameter replaces the adjusted first parameter in the near-end digital signal. Reading, creating and adjusting are preferably done by one processor.

Another embodiment of the present invention is further a communication for transmitting a near-end digital signal using a code sample consisting of a first bit using a compressed code and a second bit using a linear code. Useful in systems. The code sample represents an audio signal having a plurality of audio characteristics. The communication system also
It also transmits the far-end digital signal. In such an environment, any echo in the near-end digital signal can be reduced by adjusting the first and second bits in response to the near-end and far-end digital signals without a compression code. .

[0037] Detailed description of the preferred embodiment   Preferred embodiments of the present invention will be described with reference to the following abbreviations. ACELP Algebraic Code Excited Linear Prediction               Shape prediction) AE Audio Enhancer (Audio Enhancer) ALC Adaptive or Automatic Level Control               (Automatic level control) CD Coded Domain or Compressed Domain               Reduction domain)

[0038] CDEC Coded Domain Echo Control               Le) EFR Enhanced Full Rate (Enhanced Full Rate) ETSI European Telecommunications Standards Institute               Communication Standardization Association) FR Full Rate

GSM Global System for Mobile Communications ITU International Telecommunications Union
MR-ACELP Multi-Rate ACELP PCM Pulse Code Modulation (ITU G.711) (Pulse Code Modulation)

[0040] RPE-LTP Regular Pulse Excitation-Long Term Prediction                 Loose excitation-long-term prediction) TFO Tandem Free Operation VSELP Vector Sum Excitation Linear Prediction               Linear prediction)

Speech Synthesis Transfer Function Although many nonlinearities and heuristics are involved in speech synthesis in a decoder, the following approximate transfer function can be characterized in the synthesis process.

[Equation 1]

The codebook vector c (n) is filtered by H (z) to obtain synthesized speech. The key points to note about this generic LPC speech synthesis or decoder model for speech decoding are that the available coded parameters that can be modified to implement echo control are: c (n): codebook vector, 2. G: Codebook gain, 3. g _p: pitch gain, 4. T: pitch period 5. {A _k , k = 1 ,. ． ． , M}: LPC coefficients.

Most LPC-based vocoders use parameters similar to the above set, parameters that may be changed to the above mentioned format, or parameters related to the above mentioned format. For example, the LPC coefficients of an LPC-based vocoder may be expressed using a log-area ratio (eg GSM FR) or line spectral frequency (eg GSM EFR). Both of these formats can be converted to LPC coefficients. An example where the parameters are related to the above format is the block maximum parameter in the GSM FR vocoder. The block maximum can be considered to be directly proportional to the codebook gain in the model described by equation (1).

Thus, although the description of coding parameter modification methods is largely limited to general-purpose speech decoder models, it is relatively easy to adapt such methods to any LPC-based vocoder and possibly even other models. .

Furthermore, non-linear processing methods such as center-clipping, which are used with uncoded speech for echo control, have significantly different coding parameter representations of speech signals, so that the coding parameters are It should also be clear that it cannot be used for. Codebook vector signal c (
Even n) does not respond to center clipping due to the significant quantization involved. In many vocoders, the majority of codebook vector samples are already zero, while non-zero pulses are highly quantized. Therefore, such non-linear processing approaches are either inapplicable or ineffective.

Within the scope of this specification and the claims, the terms “linear code” and “compressed code” have the following meanings. Linear code: By linear code is meant a method that results in one coding parameter or coded sample for each sample of the audio signal. Examples of linear codes are PCM (A-law and μ-law), ADPCM (adaptive differential pulse code modulation), and delta modulation.

Compressed code: By compressed code is meant a compression method that results in less than one coding parameter for each sample of the audio signal. Typically, the compressed code results in a small set of coding parameters for each block or frame of audio signal samples. An example of compressed code is GSM
It is a linear predictive coding based vocoder such as a vocoder (HR, FR, EFR).

Controlled Overview of Coded Domain Echo FIG. 9 shows a coded domain echo control (CD) for situations where acoustic echo is present.
EC) shows a new concrete example. Communication system 10 uses near-end encoded digital signals (near end co) via network 24 using compressed codes such as any of the codes used by the codecs shown in Table 1.
ded digital signals). The compressed code is generated by the encoder 16 from the linear-audile signal generated by the near-terminated microphone 14 in the near-terminated speaker headset 12. The compressed code contains parameters such as those shown in FIG. This parameter represents an audio signal that includes multiple audio characteristics including audio level and power. The compressed code can be decoded by various decoding steps. As described below,
The system 10 includes remote end digital signals transmitted by the system 10 via a network 32.
Controls echoes in near-terminated digital signals due to the presence of. This echo is controlled by the minimal delay of the compression code parameters shown in FIG. 8 and the minimal decoding (if decoding occurs).

The remote termination digital signal using the compressed code is received by the proximity termination terminal 20, and the digital signal using the regulated compression code is passed by the proximity termination terminal 22 via the network 24 to a decoder of the regulated compression code (illustrated). (Not shown) to a remote termination handset (not shown). Note that the tailored compression code is compatible with the original compression code. In other words, when a coding parameter is changed or adjusted, we call this coding parameter a tailored compression code that is still decodable using the standard decoder corresponding to the original compression code. is there. A remote termination encoder (not shown) is provided for the linear remote termination audio signal to produce a remote termination digital signal using a compression code compatible with the decoder 18.
And is transmitted to the remote termination terminal 34 via the network 32. Decoder 18 of near-end handset 12 decodes the remote-end digital signal. As shown in FIG. 9, the echo signal from the remote termination signal will go to the encoder 16 of the near termination handset 12 via acoustic feedback.

The processor 40 performs various operations on the near-terminated compressed code and the remote-terminated compressed code. Processor 40 may be a microprocessor, microcontroller, digital signal processor, or other type of logic unit capable of arithmetic and logical operations.

For each type of codec, different coding domain echo control algorithms 44 in compressed and linear modes during TFO and non-TFO.
Are always executed by the processor 40. Partial decoder 48 is executed by processor 40 to read at least the first parameter of the parameters received at terminal 20. Another partial decoder 46 is executed by the processor 40 to generate the at least partially decoded far end signal. Decoder 48 produces an at least partially decoded near termination signal. (Note that the compression codes used by the near-end signal and the remote-end signal may be different from each other, and thus the partial decoders may also be different from each other.)

Based on partial decoding, the algorithm 44 estimates at least the amount of echo in the near-terminated digital signal by an echo likelihood signal (echo like signal).
ihood signal). The echo likelihood signal changes over time as the amount of echo depends on the remote termination voice signal. The echo likelihood signal is
Used by algorithm 44 to adjust one or more parameters read by algorithm 44. The adjusted parameter is the terminal 2
2 is written into the near-end digital signal to form a conditioned near-end digital signal transmitted to network 24. In other words, the adjusted parameters are used in place of the originally read parameters.
The partial decoders 46, 48 shown in the network ALC device are algorithms executed by the processor 40 and are codec dependent.

The partial decoder manipulates the compressed signal using the compression code. If processor 40 is implemented in a TFO environment, partial decoder 46 will decode linear code rather than compressed code. Further, in this case, the partial decoder 48 would only decode the linear code and determine the coding parameters from the compressed code without actually synthesizing the audio signal from the compressed code. Furthermore, the blocks 44, 46, 48 may be implemented as hardwired circuits.

FIG. 10 shows that the embodiment of FIG.
-WIRE-TO-2-WIRE HYBRID) indicates that it is usable for the resulting system.

The CDEC device / algorithm removes the echo effect from the near-end coded speech by directly changing the coding parameters in the bitstream received from the near-end. Decoding of the near-end signal and the far-end signal is performed in order to obtain the likelihood of the echo existing at the near-end. To determine this likelihood value, some statistics are measured from the decoded signal.

Partial Decoding The decoding of near-end signals and remote-end signals can be partial or full decoding depending on the vocoder used for the encoding and decoding operations. Good. Below are some examples of situations where partial decoding is sufficient. 1. In Code Excited Linear Prediction (CELP) vocoders, post-filtering processing is performed on the decoded signal using an LPC-based model. After this, the filtering process reduces the quantization noise. However, the post-filtering process does not have a significant adverse effect on the measurement of the statistical values needed to determine the likelihood of the echo, so that the post-filtering step can be omitted for economy.

2. In TFO in GSM networks, a CDEC device may be located between the base station and the switch (known as the A-interface) or between two switches. Since, as shown in FIG. 3, the 6 MSBs of each 8-bit sample of the speech signal correspond to the PCM code, it is possible in this situation to avoid coding the entire coded speech. . A simple table lookup is sufficient to convert 8-bit companded samples to 13-bit linear speech samples using the A-law companding table. This is a version of the audio signal without calling the appropriate decoder.
Provide an economical way to get. Although the speech signal obtained in this way contains some noise, it has been found to be sufficient for the measurement of the statistics needed to determine the echo likelihood.

Echo Likelihood Determination Assuming that some (fully or partially decoded) uncoded versions of the far-terminated signal and the near-terminated signal are available, some statistics are measured. And is used to determine the likelihood of the echo present in the near-end signal. The echo likelihood is estimated for each speech subframe, where the subframe duration depends on the vocoder used. The preferred approach is described in this section.

Suppose a simplified model of end-path is as shown in FIG. This endpath is assumed to consist of a uniform delay of τ samples and an echo return loss (ERL) λ.

In this model, s _NE (n) is the near-end uncoded signal and s _FE (n) is the far-end uncoded signal. The range of τ is known for a particular implementation of CDEC and is specified below.

[Equation 2]

This assumption is valid, since the maximum end path delay and the minimum end path delay depend mainly on speech coding, speech decoding, channel coding, channel decoding and other known transmission delays. Is. The ERL range is

[Equation 3] Is assumed to be

The process of estimating the likelihood (likelihood) of an echo uses the following variables. P _NE is the power of the current subframe of the near-end signal. P _FE (0) is the power of the current subframe of the far end signal. P _FE (m) is the power of the m-th subframe before the current subframe of the far-end signal. In other words, a buffer of past far-end subframe power values is maintained. The buffer size is B _max = [τ _max / N] so that the far-end signal subframe power up to the maximum possible end-path delay is available. Here, N is the number of samples in the subframe.

R is the ratio of near-end subframe power to far-end subframe power. ρ ₁ is the likelihood of prior echo (likelihood). ρ is the echo likelihood obtained by smoothing the prior echo likelihood (likelihood).

Estimate the echo likelihood (likelihood) for each subframe using the steps described below. For some vocoders, especially for particularly low bit rate vocoders such as GSM HR, this process may be better done on a frame-by-frame basis rather than a sub-frame basis.

The power of s _NE (n) for the current subframe is

[Equation 4] Ask as.

The power of s _FE (n) for the current subframe is

[Equation 5] Ask as.

[0067]   The ratio of near-end power to far-end power

[Equation 6] Here, B _min = [τ _min / N]. The denominator is essentially the maximum far-end subframe power measured during the expected endpath delay time period.

Shift the far-end power value in the buffer: P _FE (B _max ) = P _FE (B _max −1) ;. ． ． P _FE (1) = P _FE (0).

[0069]   Pre-echo probability (likelihood)

[Equation 7] Ask as.

Smooth the a priori echo likelihood (likelihood) using ρ = 0.9ρ + 0.1ρ ₁ ,
Obtain the likelihood of echo (likelihood).

A graph of the likelihood of prior echo (likelihood) as a function of the ratio of near-end subframe power to far-end subframe power is shown in FIG.

Coding Parameter Modification This section describes a preferred method for direct modification of coding parameters based on echo likelihood. Direct modification of each coding parameter of the general-purpose speech decoder model of FIG. 7 will be described first. Then, corresponding methods for parameter changes for standards-based vocoders are described. As an example of a standards-based vocoder, consider the GSM FR vocoder. After modifying and quantizing each parameter according to the standard, modify the appropriate parameters in the bitstream appropriately.
A preferred embodiment of this entire process is shown in FIG.

Modifying Codebook Gain The codebook gain parameter G for each subframe is reduced by a scaling factor that depends on the echo likelihood ρ for that subframe. The modified codebook gain parameter represented by G _new is given by: G _new = (1−ρ) G (4)

This parameter is then requantized according to the vocoder standard. Note that in the speech decoder model of FIG. 7, the codebook gain controls the total level of the composite signal and thus the corresponding audio signal. On the other hand, attenuating the codebook gain results in echo attenuation.

For GSM FR, the block maximum parameter X _max is directly proportional to the codebook gain parameter of the generic model of FIG. Therefore, the modified block maximum parameter is calculated as X _{max, new} = (1−ρ) X _max (5).

Then, X _{max, new} is requantized by the method specified in the standard. The resulting 6-bit value is reinserted at the proper position in the bitstream.

Codebook Vector Modification The codebook vector c (n) is modified by randomizing the position and amplitude of the pulse. The randomization of the codebook vector results in the elimination of the echo correlation properties. This has the effect of eliminating many of the "voice-like" properties of echo. Randomization is performed whenever it is determined that the likelihood of echo (likelihood) is high, preferably when ρ> 0.8. This randomization can be done using any suitable pseudo-random bit generation method.

For GSM FR, the codebook vector for each subframe is
It is determined by the RPE grid position parameter (2 bits) and 13 RPE pulses (each pulse is 3 bits). These 41 bits are replaced with 41 random bits using a pseudo-random bit generator.

Pitch Synthesis Filter Modification Pitch synthesis filters are particularly important for achieving long-term correlation of every cycle of a speech signal and for modeling harmonics of voiced speech. The model of this filter described in FIG. 7 uses only two parameters: pitch period T and pitch gain g _p . During voiced speech, the pitch period is relatively constant over some subframes or frames. Pitch gain in most vocoders is zero to 1 or slightly greater than 1 (eg 1.2 for GSM EFR)
It is within the range up to. During strong voiced speech, the pitch gain is at or near its maximum value.

If the echo is only present in the near-end signal, the voiced harmonics of that echo are generally well modeled by a pitch synthesis filter and it is detected that the likelihood of the echo is high. (Ρ> 0.8).

If both echo and near-end speech are present in the near-end signal during the frame period,
The likelihood of echo is at an appropriate level (0.5 ≦ ρ ≦ 0.8). In these situations, the encoding process will generally model the stronger of these two signals. In most cases it is reasonable to assume that the near-end speech is stronger than the echo. If this is the case, the encoding process, due to its nature, models mainly near-end speech harmonics with pitch synthesis filters,
Echo harmonics tend to model little or no modeling.

To eliminate or mask the voiced echoes, the harmonic nature of the echo is eliminated. This is achieved by changing the pitch synthesis filter parameters as follows.

The pitch period is randomized so that the long-term correlation in the echo is removed, thus eliminating the voiced nature of the echo. Such randomization is performed only when the likelihood of the echo is high, preferably ρ> 0.8.

The pitch gain is reduced to control the strength of the harmonics or the strength of the long-term correlation in the audio signal. Such attenuation of gain is preferably performed only when the likelihood of the echo is at least moderate (ρ> 0.5).

[0085]   A new pitch gain is obtained as follows.

[Equation 8]

Note that with this approach, the pitch period is not randomized during moderate echo likelihood, but the pitch gain may be attenuated so that the voiced quality of the signal is not strong. .

FIG. 14 shows the magnitude frequency response of a pitch synthesis filter having a pitch period T = 14. Dotted line is a response for a high pitch gain _(g p = 0.75), solid _line, g p = 0.
Figure 3 shows the conditions that occur when the pitch gain is attenuated. The strength of the harmonics of the audio signal and the long-term correlation can be controlled by changing this parameter in this way.

In the GSM FR vocoder, the LTP delay parameter of subframe _j , represented by N _j , corresponds to the pitch period T of the model of FIG. N _j picks up 7 bits in the bitstream and can be in the range 40 to 120. Therefore, when randomizing N _j , a random number within this range must replace N _j .

LTP of GSM FR vocoder subframe _j , represented by b _j
The gain parameter corresponds to the pitch gain g _p in FIG. 7. The modified LTP gain parameter is obtained in a manner similar to equation (6) below.

[Equation 9]

[0090] Change of LPC synthesis filter   In the general-purpose speech decoder model of FIG. 7, the LPC synthesis filter conversion function is

[Equation 10] Is. This filter provides wide area shaping for the composite signal. The magnitude frequency response of this filter is flattened by replacing the coefficients {a _k } with {β ^k a _k }, where 0 ≦ β ≦ 1. β is called the “spectral morphing factor”. In other words, the modified conversion function is

[Equation 11] Is. Note that when β = 0, the original LPC synthesis filter is transformed into an all-pass filter, and when β = 1, the original filter remains unchanged. For all values of β from 0 to 1, the magnitude frequency response of the original filter suffers some flattening, and β → 0 undergoes greater flattening. Note that the stability of the filter is maintained in this variant.

The effect of such spectral morphing on the echo is to reduce or eliminate any formant structure present in the signal. The echoes are blended or morphed to sound like background noise. As an example, FIG. 1 shows the magnitude frequency response of an LPC synthesis filter for voiced speech segments and its flattened variant for several different β values.
5 shows.

In this preferred embodiment, the spectral morphing factor β is determined as follows.

[Equation 12]

Similar spectral morphing methods are obtained for other representations of LPC filter coefficients commonly used in vocoders, such as reflection coefficient, log-area ratio, inverse sine function, and line spectral frequency. To be

For example, the GSM FR vocoder uses a log-to represent LPC synthesis filter.
Use area ratio. LAR (i), i = 1, 2 ,. ． ． , 8 for eight log-area ratios corresponding to a frame, the spectrally morphed log-area ratio is obtained using the following equation:

[Equation 13] Here, β is obtained by the equation (8).

This method spectrally flattens the magnitude frequency response of the LPC filter.
Alternatively, morph the log-area ratio to a predetermined spectrum or magnitude frequency response, such as the spectrum of background noise represented by a set of log-area ratios represented by LAR _noise (i). To do this, the appropriate morphing formula is:

[Equation 14]

The modified log-area ratio is then quantized according to the specifications in the standard. Note that such an approach to changing the log-area ratio maintains the stability of the LPC synthesis filter.

A typical example of an approach for spectral prediction of background noise and representation of filter coefficients including log-area ratios corresponding to vocoders and LPC filters is the Comfort Noise Generation Standard [5] and its And the reference to.

When line spectral frequencies are used to represent LPC synthesis filters (eg GSM EFR), an approach similar to the one for log-area ratio is also suitable. The line spectrum frequencies are f _i , i = 1 ,. ． ． , M, where M is the order of the LPC synthesis filter that is assumed to be uniform (typical). When the line spectral frequencies are uniformly spaced from 0 to 1/2 the sampling frequency, the resulting LPC synthesis filter will be all-pass (ie, flat magnitude frequency response). The set of line spectral frequencies corresponding to such a spectrally flat LPC filter is given by f _{i, flat}
= 1 ,. ． ． , M.

The spectrally morphed line spectral frequency is then obtained using the following equation:

[Equation 15] Here, β is obtained by the equation (8).

This method spectrally flattens the magnitude frequency response of the LPC synthesis filter. Alternatively, to morph the line spectral frequencies to a predetermined spectrum or magnitude frequency response, such as a background noise spectrum represented by a set of line spectral frequencies represented by f _{i, noise} , A suitable morphing formula is as follows.

[Equation 16]

The modified line spectrum frequency is then quantized according to the specifications in the standard. Note that such an approach to changing the line spectral frequency maintains the stability of the LPC synthesis filter. Suitable methods for background noise spectrum prediction and representation of filter coefficients containing line spectrum frequencies are given in the corresponding vocoder standard for comfort noise generation.

Minimum Delay Techniques Large delays in buffering, processing and transmission already exist in cellular networks without any network voice quality enhancement processing. Further network processing of the coded speech for speech enhancement will add additional delay. Minimizing this delay is important for voice quality. This section describes a new approach to minimize this delay. The case used is the GSM FR vocoder.

FIG. 8 shows the order in which the coding parameters from the GSM FR encoder are received. A simple approach involves buffering the entire 260 bits for each frame, and then processing these buffered bits for coded domain echo control. However, this causes a processing delay plus a buffering delay of about 20 ms.

This buffering delay can be minimized as follows. First, note that the entire first subframe may be decoded immediately after bit 92 is received. Therefore, this first subframe will be processed after a buffering delay of about 7.1 ms (20 ms × 92/260). Therefore, the buffering delay is reduced by about 13 ms.

Using this new low delay approach, the coded LPC synthesis filter parameters are modified based on the information available at the end of the first subframe of the frame. In other words, the entire frame is affected by the echo likelihood calculated based on the first subframe. The experiments carried out did not find any noticeable artifacts due to this "early" decision, which is particularly smoothed based on the echo likelihood being effectively based on some previous subframes and the current frame. This is because it is a converted amount.

Error Correction / Detection Bits and Updating Frame Indication Bits When applying the novel coding domain processing method for echo cancellation described herein, some or all of the bits corresponding to the coding parameters may be applied. Changed in bitstream. This can also adversely affect other error correction or detection bits that may be embedded in the bitstream. For example,
The audio encoder may embed some checksums in the bitstream for the decoder to check to ensure that error-free frames are received. Such checksums, any parity check bits, error correction or detection bits, and framing bits are updated as needed according to the appropriate standard.

Operation according to the GSM Tandem Free Operations Standard If only the coding parameters are available, partial or full decoding may be performed as described above, whereby the coding parameters are of the audio signal. Used to rebuild the version. However, when operating under circumstances such as the GSM TFO environment, additional information is available in addition to the coding parameters. This additional information is the 6 MSBs of the A-law PCM sample of the audio signal. In this case, these PCM samples may be used to reconstruct a version of the audio signal for both the far end and the near end without using coding parameters. This results in computational savings.

Those of skill in the communication arts will recognize that these preferred embodiments can be modified and varied without departing from the true spirit and scope of the invention as defined in the appended claims. Will.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of a system for voice transmission in a GSM digital cellular network.

FIG. 2 is a schematic block diagram of a system for voice transmission in a GSM network under Tandem Free Operation (TFO).

FIG. 3 is a graph illustrating voice transmission under Tandem Free Operation (TFO).

FIG. 4 is a schematic block diagram of a traditional solution to the echo problem in wired networks.

FIG. 5 is a schematic block diagram illustrating acoustic feedback from a speaker to a microphone in a digital cellular network.

FIG. 6 is a schematic block diagram of a traditional echo cancellation approach for coded speech.

FIG. 7 is a schematic block diagram of a Generic Linear Predictive Code (LPC) speech synthesis model or speech decoding model.

FIG. 8 is an illustration of the placement of coding parameters in a bitstream for GSM FR.

FIG. 9 is a schematic block diagram of a preferred form of coded domain echo control for an acoustic echo environment made in accordance with the present invention.

FIG. 10 is a schematic block diagram of another preferred form of coded domain echo control for echoes resulting from a 4-wire vs. 2-wire hybrid made in accordance with the present invention.

FIG. 11 is a schematic block diagram of a simplified end path model with flat delay and attenuation.

FIG. 12 is a graph showing a power ratio of a near-end subframe to a far-end subframe, which is a preliminary echo likelihood contrast.

FIG. 13 is a flowchart showing a priority mode of a coded domain echo control method.

FIG. 14 is a graph showing an example of the amplitude-frequency response of the pitch synthesis filter.

FIG. 15 is a graph showing an example of the amplitude-frequency response of an original LPC synthesis filter and a flattened version of such a filter.

─────────────────────────────────────────────────── ─── Continued front page    (81) Designated countries EP (AT, BE, CH, CY, DE, DK, ES, FI, FR, GB, GR, IE, I T, LU, MC, NL, PT, SE), OA (BF, BJ , CF, CG, CI, CM, GA, GN, GW, ML, MR, NE, SN, TD, TG), AP (GH, GM, K E, LS, MW, MZ, SD, SL, SZ, TZ, UG , ZW), EA (AM, AZ, BY, KG, KZ, MD, RU, TJ, TM), AE, AG, AL, AM, AT, AU, AZ, BA, BB, BG, BR, BY, BZ, C A, CH, CN, CR, CU, CZ, DE, DK, DM , DZ, EE, ES, FI, GB, GD, GE, GH, GM, HR, HU, ID, IL, IN, IS, JP, K E, KG, KP, KR, KZ, LC, LK, LR, LS , LT, LU, LV, MA, MD, MG, MK, MN, MW, MX, MZ, NO, NZ, PL, PT, RO, R U, SD, SE, SG, SI, SK, SL, TJ, TM , TR, TT, TZ, UA, UG, US, UZ, VN, YU, ZA, ZW F-term (reference) 5D045 DA20                 5J064 AA01 BA05 BA13 BB01 BB03                       BC01 BC16 BD02                 5K046 HH11 HH51 HH77 HH78

Claims (61)

[Claims] 1. A communication system for transmitting a near-terminated digital signal using a compressed code having a plurality of parameters including a first parameter, the parameter being one audio having a plurality of audio characteristics. Of the compression code can be decoded by a plurality of decoding steps, and the communication system uses the compression code to transmit the remote termination signal,
A device for reducing echoes in a near-end digital signal, the device comprising:
One processor, the processor reading at least the first parameter of the plurality of parameters in response to a digital signal at the terminal of the proximity, and performing at least one of the plurality of decoding steps of the proximity of the plurality of parameters. Performing on the terminal digital signal and the remote terminal digital signal to generate an at least partially decoded near terminal signal and an at least partially decoded remote terminal signal, and Adjusting the first parameter in response to the at least partially decoded near end signal and the at least partially decoded remote end signal to generate the adjusted first parameter; and A device for reducing echoes, which replaces a first parameter with the adjusted first parameter in the digital signal at the end of the proximity. . 2. The first parameter is a quantized first parameter and the processor adjusts the adjusted first parameter before writing the adjusted first parameter to the digital signal at the proximal end. The apparatus of claim 1, wherein the adjusted first parameter is partially generated by quantizing the adjusted first parameter. 3. The processor is responsive to the at least partially decoded near-end signal and the at least partially decoded near-end signal to the at least partially decoded near-end signal. The apparatus of claim 1, wherein a probable echo signal is generated that is indicative of the amount of echo present, and the processor adjusts the first parameter in response to the probable echo signal. 4. The feature has a spectral shape, the first parameter has a representation of filter coefficients, and the processor is responsive to a signal of likelihood of the echo to represent the filter coefficients. 4. The apparatus of claim 3, wherein is adjusted toward a magnitude frequency response. 5. Apparatus according to claim 4, wherein the representation of the filter coefficients comprises the frequencies of the spectrum of the line. 6. The representation of the filter coefficient has a ratio of log areas,
The device according to claim 4. 7. The apparatus of claim 4, wherein the magnitude frequency response corresponds to background noise. 8. The apparatus of claim 1, wherein the feature comprises an overall level of the audio signal and the first parameter comprises a codebook gain. 9. The apparatus of claim 1, wherein the first parameter comprises a vector of codebook parameters. 10. The apparatus of claim 1, wherein the feature comprises a long term correlation period and the first parameter comprises a pitch period parameter. 11. The apparatus of claim 1, wherein the feature has a long term correlation strength and the first parameter has a pitch gain parameter. 12. The apparatus of claim 1, wherein the features have a spectral shape and the first parameter has a representation of filter coefficients. 13. The apparatus of claim 12, wherein the representation of the filter coefficients comprises a ratio of log areas. 14. The apparatus of claim 12, wherein the representation of the filter coefficients comprises the frequencies of the spectrum of the line. 15. The apparatus of claim 12, wherein the representation of the filter coefficients comprises a linear predictive coded synthesis filter. 16. The first parameter corresponds to a first characteristic of the plurality of audios, and the plurality of decoding steps avoids substantially changing the first characteristic. 2. The apparatus of claim 1, having the steps of: and the processor avoiding performing the at least one decoding step. 17. The apparatus of claim 16, wherein the audio feature has power and the first feature has power. 18. The apparatus of claim 16, wherein the at least one decoding step comprises post-filtering. 19. The apparatus of claim 1, wherein the compression code comprises a linear predictive code. 20. The apparatus of claim 1, wherein the compression code comprises a regular pulse excitation versus a long term predictive code. 21. The apparatus of claim 1, wherein the compression code comprises a code-excited linear prediction code. 22. The first parameter is a sequence of first received over time.
The processor reads the series of first parameters in response to the digital signal at the near-terminated end, the processor reading the at least partially decoded near-end and far-end signal and at least The apparatus of claim 1, wherein the adjusted first parameter is generated in response to a plurality of the series of first parameters. 23. The compression code is arranged in a frame of the digital signal, the frame having a plurality of subframes, each of the subframes having the first parameter, the processor Responsive to a code of compression, reading at least the first parameter from each of the plurality of subframes, the processor including the first parameter of the adjusted first parameter in each of the plurality of subframes. The device according to claim 1, wherein 24. The processor reads the first parameter from a first one of the subframes and performs at least a plurality of the decoding steps on the near-terminated digital signal during the first subframe. And replacing the first parameter with the adjusted first parameter to achieve a smaller delay before processing a subframe subsequent to the first subframe. 23
The described device. 25. The compression code is arranged in a frame of the digital signal, the frame having a plurality of subframes, each of the subframes having the first parameter, the processor Performing at least a plurality of said decoding steps during a first one of the subframes to generate said at least partially decoded near-terminated and far-terminated signals, and setting said first parameter to said first sub-frame. Reading from a second one of the sub-frames that appear subsequent to the frame and responsive to the at least partially decoded near-end and remote-end signals and the first parameter, the adjusted first 2. The apparatus of claim 1, generating a parameter and replacing the first parameter of the second subframe with the adjusted first parameter. 26. A communication system for transmitting a near-terminated digital signal having code samples, the code samples using a compression code.
, And a second bit using a linear code, the sample of the code representing an audio signal, the audio signal having a plurality of audio features, and the system also includes a remotely terminated digital signal. An apparatus for reducing echoes in a near-end digital signal without decoding the compression code in the near-end digital signal without encoding the compression code in a system for transmitting signals, the apparatus comprising: Is an apparatus for reducing echoes comprising adjusting the first bit and the second bit in response to the near-end digital signal and the remote-end digital signal. 27. An apparatus for reducing echoes in a near-end digital signal without decoding a compression code, the first bit and the second bit in response to the near-end digital signal and the remote-end digital signal. 27. The apparatus of claim 26, comprising a processor that adjusts the bits of the. 28. The apparatus of claim 26, wherein the linear code comprises a pulse code modulation (PCM) code. 29. The apparatus of claim 26, wherein the sample code of compression conforms to tandem-free operation of global systems for mobile communication standards. 30. The first bit comprises the two least significant bits of the sample and the second bit comprises the six most significant bits of the sample.
The described device. 31. The apparatus of claim 29, wherein the 6 most significant bits comprise a PCM code. 32. A communication system for transmitting a near-terminated digital signal using a compressed code having a plurality of parameters including a first parameter, the parameter of an audio having a plurality of audio characteristics. Represents a signal,
The compression code is decodable in multiple decoding steps, and the communication system also uses the compression code to transmit a remote-ended signal, in which the echo in the near-terminated digital signal is reduced. Wherein the method reads at least a first parameter of the plurality of parameters in response to the near-end digital signal, the plurality of decodings of the near-end digital signal and the remote-end digital signal. Performing at least one of the steps of generating at least partially decoded near-end signal and at least partially decoded near-end signal, said at least partially decoded near-end signal And adjusting the first parameter in response to the at least partially decoded remote termination signal to adjust the adjusted first parameter. Step for generating a meter, and a method of reducing echo comprises the step of replacing a first parameter which is the adjustment in a parameter of the first signal of the proximity end. 33. The first parameter is a quantized first parameter and the adjusting partially tunes the adjusted first parameter by quantizing the adjusted first parameter. 32. The method of claim 31, comprising the step of generating. 34. The adjustment to the partially decoded near-end signal in response to the at least partially decoded near-end signal and the at least partially decoded near-end signal. 32. The method of claim 31, further comprising the step of generating an echo probable signal representative of the amount of echo present, the adjusting further comprising the step of adjusting the first parameter in response to the echo probable signal. the method of. 35. The feature has a spectral shape, the first parameter has a representation of a filter coefficient, and the adjustment is responsive to a signal of likelihood of the echo to adjust the filter coefficient to a magnitude frequency response. 34. The method of claim 33, comprising adjusting towards. 36. The method of claim 34, wherein the representation of the filter coefficients comprises the frequencies of the line spectrum. 37. The method of claim 34, wherein the representation of the filter coefficients comprises a ratio of log areas. 38. The method of claim 34, wherein the magnitude frequency response corresponds to background noise. 39. The feature comprises an overall level of the audio signal,
32. The method of claim 31, wherein the first parameter comprises a code butt gain. 40. The method of claim 31, wherein the first parameter comprises a vector of codebook parameters. 41. The method of claim 31, wherein the feature comprises a long term correlation period and the first parameter comprises a pitch period parameter. 42. The method of claim 31, wherein the feature comprises a long-term correlation strength and the first parameter comprises a pitch gain parameter. 43. The method of claim 31, wherein the feature has a spectral shape and the first parameter comprises one that represents a filter coefficient. 44. The method of claim 42, wherein the representation of the filter coefficients comprises a ratio of log areas. 45. The method of claim 42, wherein the representation of the filter coefficients comprises the frequencies of the line spectrum. 46. The method of claim 42, wherein the representation of the filter coefficients comprises a linear predictive coded synthesis filter. 47. The first parameter corresponds to a first feature of the plurality of audio features, and the plurality of decoding steps avoids at least one substantial change in the first feature. 32. Decoding steps, wherein the performing the at least a plurality of decoding steps comprises the step of avoiding performing the at least one decoding step.
The method described. 48. The method of claim 46, wherein the audio feature comprises power and the first feature comprises power. 49. The method of claim 46, wherein the at least one decoding step comprises post-filtering. 50. The method of claim 31, wherein the compression code comprises a linear predictive code. 51. The method of claim 31, wherein the compression code comprises a regular pulse excitation-long term prediction code. 52. The method of claim 31, wherein the compression code comprises a code-excited linear prediction code. 53. The first parameter comprises a series of first parameters received over time, the reading comprises reading the series of first parameters, and the adjusting comprises: 32. Generating the adjusted first parameter in response to at least partially decoded near and far end signals and in response to at least a plurality of the series of first parameters. The method described. 54. The compression code is located within a frame of the digital signal, the frame having a plurality of subframes, each having the first parameter, and the reading: Responsive to the code of compression, reading at least the first parameter from each of the plurality of subframes, the permutation comprising adjusting the first parameter to the adjusted parameter in each of the plurality of subframes. 32. The method of claim 31, comprising replacing with the first parameter. 55. The reading uses the first parameter of the first of the subframes.
The step of reading from the one of the subframes, the executing step comprising initiating performing at least a plurality of the decoding steps on the digital signal of the near termination during the first subframe, Comprising replacing the first parameter with the adjusted first parameter before processing a subframe subsequent to the first subframe to achieve a smaller delay. 53. The method according to 53. 56. The compression code is located within a frame of the digital signal, the frame having a plurality of subframes, each of the subframes having the first parameter, and the execution being the Performing at least a plurality of said decoding steps during a first one of the subframes to generate said at least partially decoded near-end and remote-end signals, said reading said first Parameter of the at least partially decoded near-end and remote-end signal and the at least partially decoded near-end and remote-end signals and Responsive to the first parameter, generating the adjusted first parameter, the permutation replacing the first parameter of the second subframe with the adjusted first parameter. Para 32. The method of claim 31, comprising the step of replacing with a meter. 57. A system for transmitting a near-terminated digital signal having code samples, the code samples being a first bit using a compression code and a second bit using a linear code. The sample of the code represents an audio signal, the audio signal having a plurality of audio features, the system also transmitting a remotely terminated digital signal, the compression code being A method of reducing echo in the near end digital signal without decoding, the method comprising: responsive to the near end digital signal and the remote end digital signal; Adjusting the second bit, the method comprising: reducing echoes. 58. The method of claim 56, wherein the linear code comprises a pulse code modulated (PCM) code. 59. The method of claim 56, wherein the compression code sample is consistent with tandem-free operation of Global Systems for Mobile Communications Standards. 60. The method of claim 56, wherein the first bit comprises the two least significant bits of the sample and the second bit comprises the six most significant bits of the sample. . 61. The method of claim 59, wherein the six most significant bits comprise a PCM code.