JP2000148189A - Speech processing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech processing device capable of more precisely performing the retrieval of a recorded voice. SOLUTION: The speech characteristic parameter of each speaker is preliminarily extracted by an extraction part 14, a speech model is further formed by a speech model forming part 15 and registered in a speech model recording part 13B. When the voice of a certain speaker is recorded, the characteristic parameter of the speech data recorded in a speech data recording part 13A is extracted by the extraction part 14, and the similarity to the registered voice model of each speaker is determined by a similarity calculation part 16, whereby the speaker is identified. At this time, the identified speaker code or the speaker code not identified are recorded together with the position information of the speech data corresponding thereto.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice processing device, and more particularly to a voice processing device capable of identifying a speaker.

[0002]

2. Description of the Related Art In recent years, an audio signal obtained by a microphone or the like is converted into a digital signal, which is recorded in, for example, a semiconductor memory. At the time of reproduction, the audio signal is read from the semiconductor memory and converted into an analog signal. A digital information recording / reproducing apparatus called a so-called digital recorder which outputs sound as a sound by a speaker or the like has been developed. JP-A-63-259700 discloses that
Such a digital audio recording / reproducing apparatus is disclosed.

In such a digital audio recording / reproducing apparatus, it is desired to further improve the operability and searchability when reproducing the recorded audio data. A proposal has been made. For example, the present applicant discloses a digital audio recording / reproducing apparatus provided with an index mark recording button for reproducing audio data in a desired range in Japanese Patent Laid-Open No. 10-3300.

[0004] The applicant of the present invention has proposed a processing control apparatus for audio data which enables recorded data transferred from a digital recording apparatus to a personal computer to be handled by a simple operation of the personal computer. 1
No. 49728.

Further, with the recent development of audio processing technology,
Speech recognition technology, speaker recognition technology, and the like are becoming practical. For example, Japanese Patent Application Laid-Open No. 8-153118 discloses an audio data processing device that enables search of audio data by applying speaker identification technology and reproduction of only the audio of a specified speaker.

[0006]

However, in the conventional speaker identification technology for speech which does not limit the utterance content, it is inevitable that erroneous identification is included. Thus, even if an attempt is made to reproduce a specified speaker, the voice of the speaker may not be able to be reproduced without omission, which has been a problem.

[0007] The present invention has been made in view of the above problems, and has as its object to provide an audio processing device capable of more accurately searching for recorded audio.

[0008]

In order to achieve the above object, a first speech processing apparatus of the present invention comprises a speech input means for inputting speech, and a feature parameter of registration speech for each speaker. A registered voice model recording means for recording as a registered voice model; a voice recording means for recording voice data inputted by the voice input means; and a feature parameter extracted from the voice data recorded by the voice recording means. Speaker identification means for performing a speaker identification process by obtaining a similarity between the speaker model and the registered voice model of each speaker; a speaker code corresponding to the speaker identified by the speaker identification means; Speaker identification data recording means for recording position information of audio data corresponding to the identification processing of the speaker identification means, wherein the speaker identification data recording means comprises: When unable to speaker identification of the audio data by the step, characterized in that recording the special speaker code indicating that the positional information and speaker of the audio data can not be identified.

In order to achieve the above object, a second aspect of the present invention is provided.
The audio processing device according to the first audio processing device,
The speaker identification unit includes a feature parameter extraction unit that extracts a feature parameter from the voice data recorded by the voice recording unit, a feature parameter extracted by the feature parameter extraction unit, and a registered voice model of each speaker. And a speaker identification unit for comparing the similarity calculated by the similarity calculation unit with a threshold for speaker recognition to identify a speaker. It is characterized by the following.

[0010] In order to achieve the above object, a third aspect of the present invention is provided.
The voice processing device according to the first or second voice processing device, further includes voiced detection means for detecting voiced data among voice data recorded by the voice recording means, and the speaker identification means. Is characterized in that speaker identification processing is performed on the sound data detected by the sound detection means.

In order to achieve the above object, a fourth aspect of the present invention is provided.
The voice processing device of the third voice processing device described above,
The sound detection means detects the sound based on sound / non-sound information recorded in a header part of the audio data recorded by the audio recording means.

In order to achieve the above object, a fifth aspect of the present invention is provided.
The audio processing device according to the first audio processing device,
When a speaker code is designated, the apparatus has a reproducing unit for reproducing voice data corresponding to the speaker code and voice data corresponding to the special speaker code.

[0013]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing a configuration of a digital audio processing device according to an embodiment of the present invention. FIG.
As shown in FIG.
A system control unit 10 that controls the entire voice processing apparatus.
, A microphone 1 for inputting external audio and the like, a preamplifier 2 for amplifying an audio signal from the microphone 1, and a power for amplifying an output audio signal that has been subjected to predetermined processing by an encoding / decoding processing unit 7 described later. An amplifier 6, a speaker 5 for outputting the amplified audio signal, and a COD circuit for removing unnecessary high frequency components and performing A / D conversion or D / A conversion on these input / output audio signals.
An EC3, an encoding / decoding processing unit (DSP) 7 for performing predetermined encoding and the like on the audio data A / D-converted by the CODEC 3, and an encoding / decoding unit under the control of the memory control unit 8.
A flash memory 13 for recording audio data appropriately processed by the decoding processing unit 7; and a feature parameter extracting unit for extracting predetermined parameters from audio data A / D converted by the CODEC 3 or decoded audio signals. 14
A speech model creation unit 15 for creating a speech model based on the data from the feature parameter extraction unit 14, a similarity calculation unit 16 for calculating the similarity based on the data from the feature parameter extraction unit 14, a similarity calculation Speaker specifying unit 17 that performs processing for specifying a speaker based on the calculation result of unit 16
A display unit 9 for displaying a predetermined state of the digital audio processing device; an operation input unit 11 including operation buttons or switches for recording and playback; and a power supply 12 for the digital audio processing device. .

The CODEC 3 is a low-pass filter 3A for removing unnecessary high-frequency components from the audio signal from the microphone 1, an A / D converter 3B for A / D converting the analog audio signal, an encoding / decoding processing unit. 7 is provided with a D / A converter 3C for D / A converting the audio data from 7 and a low-pass filter 3D for removing unnecessary high frequency components from the D / A converted audio signal.

The encoding / decoding processing unit 7 is, for example, a CELP (Code Excited Linea).
Encoding / decoding of audio is performed by an r Predictive Coding (DSP) method, and a DSP (Digital
Signal Processor).

The system control section 10 controls the operation of each section of the digital audio processing apparatus, and in this embodiment, is constituted by an 8-bit CPU. It is connected to the encoding / decoding processing unit 7 and the flash memory 13 via a memory control unit 8 composed of a 16-bit CPU, and has a CODEC 3, a display unit 9 and operation buttons for recording and reproduction or the like. An operation input unit 11 including operation switches and a power supply 12 are connected to each other.

The flash memory 13 is divided into three areas: a voice data recording section 13A, a voice model recording section 13B, and a speaker code recording section 13C. Further, the flash memory 13 may be built in the audio processing device, or may be configured to be detachable.

The feature parameter extracting unit 14 is connected to the encoding / decoding processing unit 7 and the speech model creating unit 1
5 is connected to the memory control unit 8. Further, the feature parameter extraction unit 14 is connected to the memory control unit 8 via the similarity calculation unit 16 and the speaker identification unit 17. The main operation will be briefly described. When the operator performs a recording operation using the operation input unit 11, an analog audio signal input by the microphone 1 under the control of the system control unit 10 is amplified by the preamplifier 2 and the low-pass filter 3.
Unnecessary high frequency components of the audio signal components are blocked by A. The output signal from the low-pass filter 3A is A /
It is converted into a digital signal by the D converter 3B.

Thereafter, the encoding / decoding processing section 7 performs A / D
The digital signal from the converter 3B is subjected to an encoding process. The encoded data obtained by this encoding process is stored in the area of the audio data recording unit 13A of the flash memory 13 via the memory control unit 8.

During this series of operations, the memory control unit 8
The input / output operation of a signal exchanged between the flash memory 13 and the encoding / decoding processing unit 7 is controlled. Also,
The flash memory 13 records header information in addition to the encoded data output from the memory control unit 8. The header information includes, for example, Japanese Patent Application Laid-Open No. 9-218 by the present applicant.
As disclosed in Japanese Unexamined Patent Publication No. 694, there is mentioned sound / silence information.

When the operator performs a reproducing operation by the operation input unit 11, the encoded data is read from the flash memory 13 via the memory control unit 8 under the control of the system control unit 10, and the memory control unit 8 Is supplied to the encoding / decoding processing unit 7 via the, and decoded data is created.

This decoded data is the D / A of CODEC3.
The converter 3C converts the analog sound signal into an analog sound signal, and the analog sound signal is cut off in the low-pass filter 3D of unnecessary high frequency components among its frequency components. And
The signal is amplified by the power amplifier 6 and output from the speaker 5 as a reproduction signal.

Next, the operation of speaker registration for speaker identification processing in the digital voice processing apparatus will be described. It is assumed that this speaker registration is performed prior to the above-described recording operation.

FIG. 2 is a flowchart showing the flow of speaker registration. When the operator performs a speaker registration operation via the operation input unit 11, the speaker registration voice data is input to the feature parameter extraction unit 14 (step S1). At this time, the speaker registration voice data may be input by the speaker from the microphone 1, or the speaker registration voice data may be recorded in advance, and the recorded data may be input. .

The feature parameter extracting unit 14 extracts the input registration voice data into a representation format suitable for speaker identification, for example, feature parameters such as pitch and cepstrum (step S2). Next, the time-series data of the feature parameters is input to the speech model creation unit 15, and a standard pattern of the feature parameters is created as a speech model (step S3). Then, the created audio model is recorded in the audio model recording unit 13B of the flash memory 13 via the memory control unit 8 (Step S4).

The above-described speaker registration operation is performed for each speaker to be recorded.

Next, a process for identifying a speaker with respect to recorded data, that is, audio data recorded in the audio data recording unit 13A will be described.

Here, the speaker is identified with respect to the audio data recorded in the audio data recording unit 13A. Generally, the amount of calculation for speaker identification processing is enormous.
This is because it is difficult for a small and inexpensive voice processing device to perform speaker identification processing of voice from a microphone during recording.

As described above, by identifying the speaker with respect to the audio data recorded in the audio data recording unit 13A, the timing of executing the speaker identification processing is increased. For example, this processing may be automatically performed immediately after the recording is completed, or may be performed when the operator performs a speaker identification operation via the operation input unit 11.

FIG. 3 is a flowchart showing a speaker identification process. First, the value of the variable f indicating the frame number of the audio data is set to "0" (step S5). Subsequently, "1" is added to the value of f (step S6). And
It is determined whether or not the sound of the frame corresponding to the value of the variable f is sound (step S7). This determination method may be, for example, a method disclosed in Japanese Patent Application Laid-Open No. 9-281987 by the present applicant, or a method disclosed in Japanese Patent Application Laid-Open No. 9-218 described above.
As disclosed in Japanese Unexamined Patent Application Publication No. 694, sound / silence information recorded in a header in advance may be used.

If the determination in step S7 is yes, the address information of the audio data recording unit 13A in which the audio data of the current frame is recorded is stored in the speaker code recording unit 13.
Recorded in C (step S8). Next, the voice of the current frame is input to the feature parameter extraction unit 14 to extract a predetermined feature parameter (step S9).

The similarity calculator 16 calculates the similarity between the voice model of each speaker recorded in the voice model 13B and the above-mentioned characteristic parameter (step S10). Next, the speaker whose similarity is higher than the speaker identification threshold is specified as the speaker of the current frame voice (step S11).

At this time, for a plurality of speakers, when the similarity becomes larger than the speaker recognition threshold value or when the similarity becomes larger than the speaker recognition threshold value, If there is no speaker, the speaker at that time cannot be specified. Then, a speaker code corresponding to the speaker or a code indicating that the speaker could not be specified is recorded in the speaker code recording unit 13C so as to correspond to the information recorded in step S8 (step S12).

Next, it is determined whether or not the file ends (step S13). If no, the process returns to step S6 to repeat the process for the next frame. If yes, the process ends. If the determination in step S7 is no, no processing is performed for the frame and the process returns to step S6 to repeat the processing for the next frame.

As described above, the speaker identification processing is performed only for the sound section, since the silent section does not have the personality of the speaker, it is unnecessary for calculating the similarity, and the sound section is not necessary. This is because if only the speaker identification processing is performed, identification can be performed with higher accuracy.

After the above-described speaker identification processing, the operator performs a speaker selection operation and a reproduction operation via the operation input unit 11 so that only the selected speaker and the voice of the unspecified part can be obtained. Enables playback. For example, when playing back recorded data such as an interview or a roundtable, only a specific speaker can be played back, and the content can be grasped much faster.

With current speaker identification technology, it is difficult to always accurately identify a speaker, and erroneous identification often occurs.
In such a situation, as described above, when a speaker cannot be identified, the speaker is not identified, so that the speaker is not reproduced even if the speaker is specified. Can be prevented.

[0039]

As described above, according to the present invention, it is possible to provide a voice processing apparatus capable of searching for a recorded voice more accurately.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a digital audio processing device according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a flow of speaker registration in the digital audio processing device of the embodiment.

FIG. 3 is a flowchart showing speaker identification processing in the digital audio processing device of the embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Microphone 3 ... CODEC 3A ... Low-pass filter 3B ... A / D converter 3C ... Low-pass filter 3D ... D / A converter 7 ... Encoding / decoding processing part 8 ... Memory control part 10 ... System control part 11 ... Operation Input unit 13 Flash memory 13A Audio data recording unit 13B Audio model recording unit 13C Speaker code recording unit 14 Feature parameter extraction unit 15 Audio model creation unit 16 Similarity calculation unit 17 Speaker identification unit

Claims (5)

[Claims] 1. A voice input means for inputting voice, a registered voice model recording means for recording a feature parameter of a voice for registration as a registered voice model for each speaker, and a voice data input by the voice input means. Voice recording means for recording; and a feature parameter is extracted from the voice data recorded by the voice recording means, and a similarity between the feature parameter and the registered voice model of each speaker is obtained to perform speaker identification processing. Speaker identification data recording for recording speaker identification means, a speaker code corresponding to the speaker identified by the speaker identification means, and position information of voice data corresponding to the identification processing of the speaker identification means Means, the speaker identification data recording means, when the speaker identification of the voice data could not be performed by the speaker identification means, the voice data And a special speaker code indicating that the speaker could not be specified. 2. The speaker identification unit includes: a feature parameter extraction unit configured to extract a feature parameter from voice data recorded by the voice recording unit; a feature parameter extracted by the feature parameter extraction unit; A similarity calculating means for calculating a similarity with the registered voice model of the speaker, and a speaker identification for identifying the speaker by comparing the similarity calculated by the similarity calculating means with a threshold for speaker recognition. The voice processing device according to claim 1, comprising: 3. The apparatus according to claim 1, further comprising: a voice detection unit configured to detect voice data in the voice data recorded by the voice recording unit; wherein the speaker identification unit includes a voice detected by the voice detection unit. 3. The speech processing device according to claim 1, wherein speaker identification processing is performed on the data. 4. The sound detecting device according to claim 3, wherein said sound detecting means detects the sound based on sound / no sound information recorded in a header portion of the sound data recorded by said sound recording means. Voice processing device. 5. A reproduction means for reproducing voice data corresponding to the speaker code and voice data corresponding to the special speaker code by designating the speaker code. 2. The audio processing device according to 1.