JP2007228028A - Evaluation method of acoustic echo canceller, and storage medium for use therein - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a convenient acoustic echo canceller (AEC) evaluation method not requiring a special device, and to provide a medium for use therein. <P>SOLUTION: The acoustic echo canceller is evaluated by a step of preparing near end data, remote end data, and synthetic voice data compounding near end data and echo voice data obtained from the remote end data through a predetermined acoustic system as temporally synchronized data. Consequently, a convenient AEC evaluation method not requiring a special device for evaluation and a medium for use therein are obtained. <P>COPYRIGHT: (C)2007,JPO&amp;INPIT

Description

  The present invention relates to a method for evaluating an acoustic echo canceller and a storage medium used therefor.

  The acoustic echo canceller is used to remove echo generated between a speaker and a microphone of a mobile phone, and is an indispensable filter particularly for a telephone equipped with a hands-free system. In recent years, the demand for acoustic echo cancellers has increased with the spread of mobile phones.

  Here, the function of a general acoustic echo canceller (hereinafter also referred to as AEC) will be described. The AEC is provided in the telephone 10 such as a cellular phone in the functional explanatory diagram of FIG. Usually, the sequence in which the acoustic echo canceller operates is the following five stages. First stage: Far-end speaker's voice 1 (black arrow) is input to the telephone from the telephone network and input to the acoustic echo canceller. Second stage: The far-end speaker's voice is copied into the acoustic echo canceller and output to the speaker. Since the buffer capacity that can be used to hold data is limited, the old data is usually discarded first.

  Third stage: The far-end speaker's voice output from the speaker echoes and flows into the microphone, and is mixed with the near-end talker's voice (open arrow). The far-end speaker's voice is attenuated in amplitude by echoes, causing a time delay. The degree of attenuation / delay varies greatly depending on the conditions of the near-end side room acoustic system 20 (for example, indoor / outdoor / in a train).

  Fourth stage: Near-end speaker's voice and far-end speaker's voice 2 are mixed and input to the acoustic echo canceller. The acoustic echo canceller estimates the far-end talker's voice 2 based on the far-end talker's voice copied in the second step, and subtracts it from the mixed voice to extract only the near-end talker's voice. To do. Fifth stage: The extracted near-end speaker's voice is output to the telephone network.

  The performance of the acoustic echo canceller depends on whether or not the “estimation of far-end speaker's speech” in the fourth stage is appropriately performed. Therefore, when implementing an acoustic echo canceller, it is essential to evaluate its performance.

  Also, for example, when using a mobile phone in a car, etc., there is a problem that the quality of the acoustic echo canceller is low, so that the call quality is lowered. The acoustic echo canceller was also required to be developed quickly.

  However, the acoustic echo canceller has to be mounted as described above and individually evaluated for performance. Conventionally, since this has been performed individually, there has been a problem that it takes time for evaluation. In order to solve this problem, Patent Document 1 stores a voice as a voice data for evaluation in a database, and a voice generation simulator uses the voice for each evaluation to improve reproducibility.

  The outline of the prior art disclosed in Patent Document 1 will be described with reference to the block diagram of FIG. 2. In order to reduce the time required for tuning the hands-free device, the evaluation voice is stored in the database in advance. Repeated use improves the reproducibility of evaluation of hands-free devices. For this purpose, three databases are prepared: a near-end speech database, a far-end speech database, and a database representing the characteristics of the room acoustic system (equivalent to the near-end side room acoustic system 20 in FIG. 1). By using these to evaluate an acoustic echo canceller (AEC), it is intended to improve the tuning efficiency of the hands-free device.

JP 2004-281419 A

  In the evaluation of the acoustic echo canceller, it is necessary to input near-end and far-end sounds (human voice, surrounding sounds, etc.) as described above. However, in the prior art, the near-end / far-end audio data is managed as separate data, and in order to improve reproducibility, a special device that inputs them in synchronization is required.

  A little more detail on the above reasons. In the prior art disclosed in Patent Document 1, that is, the far-end sound source database of FIG. 2, the voice generated as the far-end side is stored in the database. By passing it through a telephone network simulator, it is generated as far-end speech data and input to an acoustic echo canceller that is an evaluation target. The near-end room acoustic database is a database of acoustic characteristics for each near-end room acoustic system (in a train, in a room, etc.), which is stored in a near-end room simulator. Reproduce the echo of the far-end voice. This is added to the sound of the near-end sound source database and input to the acoustic echo canceller to be operated.

  In this prior art, the far-end sound source, the near-end sound source, and the near-end room acoustic system characteristics are stored in a database, and both the far-end side (circuit side) and the near-end side go through a dedicated simulator. By reproducing the voice and operating the acoustic echo canceller, the voice data for evaluation can always be prepared with the same quality.

  However, in order to realize this, it is necessary to align the timings of inputting the far-end side sound and the near-end side sound to the acoustic echo canceller. Although this side is not specifically disclosed in the above-described prior art, a special device is required to synchronize the timing of input near-end / far-end audio data and input it to the acoustic echo canceller. The problem arises.

  The present invention intends to provide a simple AEC evaluation method that solves the problem of the prior art that a special apparatus is required for evaluation, and a medium used therefor. Other effects of the present invention will be described later.

  The acoustic echo canceller evaluation method of the present invention synthesizes near-end speech data, far-end speech data, the near-end speech data, and reverberation speech data obtained from the far-end speech data via a predetermined acoustic system. Preparing the synthesized voice data as time-synchronized data, inputting the far-end voice data and the synthesized voice data to the acoustic echo canceller, and performing echo cancellation processing on the synthesized voice data And outputting the processed audio data.

  In other words, the present invention uses near-end voice data, far-end voice data, and synthesized voice data prepared as time-synchronized data, so that the near-end / far-end voice for input is used. The present invention solves the above-described problem that a special device is required to synchronize the timing of data or the like and input it to an acoustic echo canceller.

    According to the present invention, a simple AEC evaluation method that does not require a special device for evaluation and a storage medium used therefor can be obtained.

  The gist of the first embodiment of the present invention will be described with reference to FIG. FIG. 3 shows the outline of the audio data file used in the first embodiment of the present invention and the relationship between the input and output to the AEC 10. The input data 1 includes far-end voice data, reverberant voice data, and near-end voice data as data digitized together. These audio data are sampled at a predetermined rate based on the sampling theorem. Here, the echo sound data is obtained by adding the near-end side room acoustic system characteristics to the far-end sound data. Note that the input data 1 is in one file format. File format data refers to a collection of data recorded on a storage medium such as a hard disk, floppy (registered trademark) disk, or CD-ROM. In the present invention, any type of storage medium can be used. Here, one file format means a single file format and may be included in a file having a plurality of numbers (names). If there is a single file with the same number (name), there is a merit that the data recorded and saved there can be managed more easily. Therefore, one normal file with the same file name. Of course, the data may be recorded and stored in the memory.

  Here, the near-end room acoustic system characteristic means an acoustic system characteristic of an environment where a speaker who listens to far-end voice data is present, for example, in an automobile, an office environment, or a gymnasium. Therefore, in the case of room acoustic system characteristics, the room is different from the room generally called and should also be called a predetermined space. Therefore, strictly speaking, it should be simply a predetermined acoustic system characteristic, but it is called a room acoustic system characteristic in accordance with conventional usage.

  According to FIG. 3, the near-end voice data and the echo voice data are synthesized at the synthesis point 11 and put into the AEC 10 as synthesized voice data. Since near-end voice data and reverberation voice data are digital data, when synthesizing as will be described later, analog data is used by using a DA converter (not shown) or the like, and after being physically synthesized, AD (not shown) Digital data is input to the input terminal of the AEC 10 using a converter or the like.

  In AEC10, only the voice of the near-end speaker is extracted by subtracting from the synthesized voice data based on the digital far-end voice data obtained from other input terminals, that is, the voice data of the far-end speaker. Output as post-processing output. As the output data file, post-processing output data, that is, assumed near-end audio data and pre-processing data output from the AEC are obtained. The AEC is evaluated by comparing the post-processing output data with the actual near-end voice data. At this time, output data before processing is also sometimes referred to. Specific embodiments will be described later.

  The data structure of this input data file will be described in a little more detail using Table 1. The data files in Table 1 are far-end voice data, reverberant voice data, and near-end voice data having the following three types of data. Here, the echo sound data is obtained by mixing the far-end sound data with specific room acoustic system characteristics. The echo sound data may be obtained by either actual measurement or simulation, but it is needless to say that actual measurement is more preferable. As described above, all are digital data.

  In more detail, the reverberant sound data in Table 1 is obtained by digitally converting data measured over a predetermined time in a field having various room acoustic system characteristics with respect to specific far-end sound data A1 (t). is there. For example, file 1 is an ordinary passenger car, a so-called ordinary car, file 2 is a large bus, file 3 is a train, and so on. In this case, f1 (t) indicates a room acoustic system of a normal vehicle, and A1 · f1 (t) indicates reverberation sound data affected by the room acoustic system of a normal vehicle. Further, f2 (t) indicates the indoor acoustic system of the large bus, and A1 · f2 (t) indicates reverberant sound data that is affected by the indoor acoustic system of the large bus. Hereinafter, it is the same, and it is the data of what was measured by various room acoustic systems.

  Note that near-end voice data also uses specific data in this case. The reason why the far-end voice data and the near-end voice data are not changed is that this input file is to be viewed from the viewpoint of what happens with various changes in the room acoustic system as an AEC evaluation. Of course, as will be described later, the file structure can be changed in various ways.

本発明の第１の実施形態に用いるデータを説明する表である。

It is a table | surface explaining the data used for the 1st Embodiment of this invention.

  FIG. 6 is an example showing an example of specifically evaluating AEC using this input data. Here, the input data file is recorded by digitally converting each sound or the like. The input data is stored in a storage device 22 of a personal computer (hereinafter referred to as PC) 20 via an input terminal (not shown). Far-end voice data (ID1), echo voice data (ID2), and near-end voice data (ID3) are output from an output terminal (not shown) of the PC 20.

  Among these, the far-end voice data (ID1) is input to the AEC 10 as it is. The echo sound data (ID 2) and the near-end sound data (ID 3) enter the R and L terminals of the synthesizer 13 via the digital-analog converter (DAC) 12, respectively. The synthesizer uses, for example, R and L terminals of a stereo speaker, and the output side is made common so that the analogized data is voltage-synthesized and output as one analog data, and an analog-digital converter (ADC) 14 is input to the AEC 10 as synthesized voice data (ID4) obtained by synthesizing them.

  Here, the stereo speaker is introduced as an example of the synthesizer 13 because it has only two input terminals and is convenient for synthesizing these sounds. Basically, anything may be used. That is, after the two data are DA-converted to analog data, they can be electrically inserted into the same terminal to make one analog data, and then added again to the AEC 10 through the ADC 14. It suffices if it has two terminals and can take it out as an output of one terminal. Here, FIG. 5 shows an image diagram of combining two input data. As can be seen from FIG. 5, data is added at the addition point. 3 performs the function of the addition point in FIG. 5 and the combiner 13 in FIG.

  As a result, processed audio data (OD1) is obtained from the AEC 10 as output data. This processed audio data should output only the first near-end audio data (ID3) when the AEC 10 exhibits ideal characteristics. Therefore, it may be compared with this 1D3. Although the near-end voice data (ID3) to be compared is not visible, digital data in the input file may be used for this. Specifically, by setting the acoustic echo canceller evaluation environment in the PC 20, data in a state before performing the acoustic echo canceller processing is stored, and an output file can be obtained by outputting simultaneously with the processed data. . For example, since the AEC 10 normally incorporates a program for synchronizing data, it may be used. More specifically, when evaluating the acoustic echo canceller, a DSP evaluation board (not shown) may be used and controlled by a PC in which emulator software for controlling the DSP evaluation board is installed.

  Further, as one means for evaluating the AEC 10, it is conceivable to compare the synthesized voice data before processing with the voice data after processing, and this is also returned to the PC 20 via an input terminal (not shown). By making these comparison data files in the PC 20, an AEC evaluation data file as digital data can be obtained.

  As described above, in Embodiment 1, AEC can be easily evaluated using a simple configuration. Therefore, in the case of the prior art, a special device is required to synchronize the timing of the input near-end / far-end audio data and input it to the acoustic echo canceller, and this special device performs evaluation. Since this is realized by software such as a simulator or an emulator, this causes a problem that the evaluation environment becomes complicated, but these are also solved.

  Next, a second embodiment of the present invention will be described with reference to FIG. 4, FIG. 7, Table 2, and the like. In the second embodiment, simply speaking, the synthesized voice data is obtained in advance using the PC 20, DAC 12, stereo speaker 13, ADC 14, etc. used in the first embodiment, and this is obtained as shown in FIG. If it is a digital data file, it can be easily evaluated for various AECs.

  In order to avoid duplication of explanation, the description will focus on the points different from the first embodiment. As described above, in the second embodiment, synthesized voice data is used as an input file instead of reverberant voice data as described above. The difference is that no work is done. The structure of this data file is shown in Table 2. Therefore, in FIG. 4, the synthesized speech data is input to the AEC 10 as it is. Further, in FIG. 7, the synthesizer 12, the DAC 11, and the ADC 13 are not necessary. Therefore, there is an effect that various AECs can be evaluated more easily.

本発明の第２の実施形態に用いるデータを説明する表である。

It is a table | surface explaining the data used for the 2nd Embodiment of this invention.

  As described above, as the audio data, digital data sampled at a predetermined rate is prepared based on the sampling theorem. In order to prepare reverberant sound data, in the case of white noise or the like, it is created using the sound generation function of a general waveform editing tool. In addition, when using actual voice (such as a human voice), a recorded analog signal sampled at a predetermined sampling rate is used.

  In addition, for comparison between the output after AEC processing and the near-end audio data of the input file, devices that output stereo data in synchronization with the left and right of stereo data are generally widespread, so there is no need to prepare a special device. Synchronous output is possible. If you use the waveform editing tool to display in stereo (L side waveform and R side waveform are displayed separately), it is easy to compare and check.

  In the first and second embodiments, the measurement data (echo acoustic data) obtained by changing the room acoustic system in various ways is used as the data. However, the echo sound data and the near-end sound data obtained by one measurement are used. It is also possible to simulate different room acoustic systems by varying the time synchronization of. In the evaluation of the acoustic echo canceller, a plurality of sets of audio data (near-end / far-end audio) in which the timings of the near-end audio and the far-end audio are slightly shifted are used. In this case, in the prior art, timing variations cannot be included in the evaluation data itself, and the evaluation data is divided into near-end voice data / far-end voice data / timing data. However, this problem is also solved in the present invention.

It is a figure explaining the function of an acoustic echo canceller (AEC). It is a figure of a prior art. It is explanatory drawing explaining the summary of the 1st Embodiment of this invention. It is explanatory drawing explaining the summary of the 2nd Embodiment of this invention. It is explanatory drawing explaining a voice synthesis | combination. It is a block diagram of a 1st embodiment of the present invention. It is a block diagram of the 2nd Embodiment of this invention.

Explanation of symbols

1 Input data 2 Output data 10 Acoustic echo canceller (AEC)
11 synthesis point 12 DAC
13 Synthesizer 14 ADC
20 PC
21 CPU
22 Storage device 23 Display device

Claims (7)

An evaluation method for an acoustic echo canceller,
Near-end audio data and
Far-end voice data,
Synthesized speech data obtained by synthesizing the near-end speech data and the reverberation speech data obtained from the far-end speech data via a predetermined acoustic system;
Preparing as time-synchronized data,
Input the far-end voice data and the synthesized voice data to the acoustic echo canceller,
Outputting post-processing audio data obtained by performing echo cancellation processing on the synthesized audio data;
A method for evaluating an acoustic echo canceller, comprising: An acoustic echo canceller evaluation method according to claim 1,
The step of preparing the synthesized speech data obtained by synthesizing the near-end speech data and the reverberation speech data obtained from the far-end speech data via a predetermined acoustic system as time-synchronized data,
Data represented in one file format in which the near-end voice data and the synthesized voice data obtained by synthesizing the far-end voice data via reverberant voice data via a predetermined acoustic system are temporally synchronized. The method for evaluating an acoustic echo canceller according to claim 1, wherein An acoustic echo canceller evaluation method according to claim 1,
The step of preparing the synthesized speech data obtained by synthesizing the near-end speech data and the reverberation speech data obtained from the far-end speech data via a predetermined acoustic system as time-synchronized data,
Preparing the near-end audio data, the far-end audio data, and the reverberant audio data as data represented in one file format that is temporally synchronized;
The method for evaluating an acoustic echo canceller according to claim 1, further comprising a step of synthesizing the near-end speech data and the reverberation speech data into synthesized speech data.   4. The echo sound data according to claim 1, wherein the echo sound data is generated by performing data processing so as to cause a time delay of the predetermined room acoustic system when the echo sound data is generated. The method for evaluating an acoustic echo canceller according to any one of claims 1 to 3.   5. The predetermined acoustic system according to claim 1, wherein the predetermined acoustic system is prepared as a plurality of pieces of data obtained by editing a time synchronization relationship with the near-end audio data, the far-end audio data, and the reverberant audio data according to a predetermined rule. 5. The acoustic echo canceller evaluation method according to claim 1, further comprising: A storage medium for evaluation recording evaluation data used in an evaluation method of an acoustic echo canceller,
Near-end audio data and
Far-end voice data,
Synthesized speech data obtained by synthesizing the near-end speech data and the reverberation speech data obtained from the far-end speech data via a predetermined acoustic system;
Is recorded in the storage medium as the evaluation data synchronized in time,
The far-end voice data;
The synthesized speech data;
Is input to the acoustic echo canceller to evaluate the performance of the acoustic echo canceller. A storage medium for evaluation recording evaluation data used in an evaluation method of an acoustic echo canceller,
Near-end audio data and
Far-end voice data,
The far-end voice data is recorded in the storage medium as the evaluation data synchronized in time with the reverberant far-end voice data obtained through the predetermined acoustic system,
The far-end voice data;
Synthesized voice data obtained by synthesizing the near-end voice data and the echo voice data;
Is input to the acoustic echo canceller to evaluate the performance of the acoustic echo canceller.

Images