JPS5982608A - System for controlling reproducing speed of sound - Google Patents

Abstract

PURPOSE:To obtain a reproduced sound of high quality by dividing sound data into frames with fixed length, detecting that each frame is a silent section, a silent consonant section, a steady vowel section, or a non-steady section, and then thinning or interpolating the frame in accordance with the detected result. CONSTITUTION:A sound signal is transferred from an input buffer 3 to a silence detecting part 4 in each frame. The detecting part 4 calculates the power of a signal corresponding to the frame and the number of times of zero-crossing, and when the value is a threshold or more, regards the frame as the non-steady section (c) or the steady vowel section (d) and transfers the signal to a steady discriminating part 5. When less than the threshold, the frame is regarded as the silent section (a) or the silent consonant section and transferred to an interpolation/thinning part 7. The discriminating part 5 claculates the coefficient of self-correlation of the sound signal, and when the maximum value is larger or less than the threshold, regards the signal as the section (d) and transfers the signal to a reference period extracting part 6 or regards the signal as the section (c) and transfers the signal to an output buffer 9. The extracting part 6 calculates the reference period T of the voice signal from the coefficient providing said maximum value and transfers the period T together with the sound data to the interpolation/thinning part 7. The part 7 executes interpolation or thinning corresponding to the reproducing speed in each period T.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field of the Invention The present invention relates to a sound playback speed control method that can increase or decrease the playback speed without changing the frequency of the playback sound when playing back recorded sound. It is something.

Conventional technology and problems In tape recorders, etc., when the playback speed is changed,
The frequency of the reproduced sound changes as the reproduction speed changes, making it very difficult to hear. In order to eliminate this drawback, audio is divided into frames every several tens of milliseconds, and the audio is thinned out or interpolated in units of frames, thereby making it possible to reproduce sound without changing the frequency of the reproduced sound. Some proposals have been made that allow the speed to be increased or decreased, but
Since click sounds are generated at discontinuous points between adjacent frames after thinning or interpolation, there is a drawback that the clarity of the reproduced sound deteriorates.

In order to improve the above-mentioned drawbacks, a method has also been proposed in which the cross-correlation coefficient of the frame connection part is calculated and the connection timing is finely adjusted so that frames can be connected continuously.
The length and position of frames to be thinned out or interpolated,
Because it is fixedly determined regardless of the nature of the voice, it has the disadvantage that the clarity of phonemes that change state in a short period of time, such as plosives, deteriorates.

In addition, a method has also been proposed in which only the silent portion is expanded or contracted, but with this method, it is virtually impossible to control the speed of the audio portion.

Purpose of the Invention The present invention improves the above-mentioned drawbacks, and its purpose is to increase or decrease the reproduction speed without changing the frequency of the reproduced sound and without degrading the quality of the reproduced sound. The purpose is to do so. Examples will be described in detail below.

Embodiments of the Invention First, the present invention will be described in detail with reference to the audio signal waveform diagram of FIG. In the figure, a represents a silent section, b a stationary unvoiced consonant section, C a non-stationary section, and d a stationary vowel section.

In the silent section a1, the steady unvoiced consonant section, the amplitude of the waveform is small as shown in the figure, so even if thinning or interpolation is performed with an appropriate section length, the clarity of the reproduced sound will hardly be affected. In addition, in the stationary vowel interval d, similar waveforms are repeated every basic period T of the voice as shown in the figure, so by thinning out or interpolating at the basic period T, the frequency components of the voice can be changed. It is possible to expand and contract the reproduced sound without causing any waveform discontinuities.

However, in the non-stationary interval C, the properties of the speech waveform change rapidly, so if thinning or interpolation is performed, the characteristics of the phoneme will be lost, and in the case of short phonemes, , since the phoneme itself is missing, the intelligibility for the player deteriorates.

For the reasons mentioned above, the present invention performs thinning or interpolation with an appropriate interval length in the silent interval a1 and the steady unvoiced consonant interval, and also performs thinning and interpolation in the unsteady interval C. First, in the stationary vowel interval d, the fundamental period Tf of the voice is thinned out or interpolated as a unit, without changing the frequency components of the voice and without causing almost any discontinuity. This allows the playback speed to be increased or decreased.

FIG. 2 is a block diagram of an embodiment of the present invention, in which 1 is an analog audio signal input terminal, 2 is an AD converter, 3 is an input buffer memory, 4 is a silence detection section, 5 is a stationarity determination section,
Reference numeral 6 designates a basic period extraction section, 7 an interpolation/decimation section, 8 an output buffer memory, 9 a DA converter, and IO an output terminal.

An analog audio signal from an input terminal 1 is digitally encoded by an AD converter 2 and then stored in a buffer memory 3 in accordance with a clock signal having a constant period t. Audio data is read out from the input cover 7a 3 every l frame and transferred to the silence detection section 4.

It is assumed that one frame contains, for example, 32 milliseconds worth of audio data. The silence detection unit 4 calculates the power and zero crossing number of the audio signal corresponding to the frame based on the audio data in the frame, and if these are greater than a predetermined threshold, the frame is determined to be sound. , that is,
It is assumed that this is an unsteady section C or a steady vowel section d, and the audio data is transferred to the stationarity determining section 5. If the value is less than or equal to a predetermined threshold, it is assumed that the frame is silent, that is, a silent section a or a silent consonant section, and the audio data in the frame is transferred to the interpolation/decimation unit 7.

The stationarity determination unit 5 determines, based on the audio take within the frame,
Calculate the autocorrelation coefficient of the audio signal corresponding to that frame, and if the thickest value is larger than a predetermined threshold, consider that the frame has periodicity, that is, it is a stationary vowel interval d, The audio data within the frame is transferred to the fundamental period extraction section 6. If the maximum value of the autocorrelation coefficient is less than or equal to a predetermined threshold, the frame is considered to be in the non-stationary section C, and the audio data in the frame is added to the output buffer memory 9.

The fundamental period extraction section 6 calculates the fundamental period T of the audio signal from the coefficient that inputs the thickest value of the autocorrelation coefficient, and transfers the calculated fundamental period T to the interpolation/decimation section 7 together with the audio data.

The interpolation/decimation section 7 performs interpolation or thinning on the audio data transferred from the basic period extraction section 6, using the basic period T as a unit, and performs interpolation or thinning according to the desired playback speed. Output audio data buffer memory 8
For example, if you want to reduce the playback speed to 1/2, as shown in Figure 3 (A), one frame is
Divide into an interval (1) to (n-1) and a remainder interval (n) with the basic period T as a unit, and interpolate each interval by repeating each interval twice. The remaining section (n) is outputted only once as is. In addition, for example, when doubling the playback speed, one frame is divided into sections (1) to (n-1) whose unit is the basic period T and the remainder section (
n), and for interval (1) to (n-1),
Every other section is thinned out and output, and the remaining section (n) is output as is.

Furthermore, regarding the audio data directly transferred from the silence detection unit 4, the interpolation/decimation unit 7 performs interpolation or thinning in units of a certain length (for example, 5 milliseconds) in accordance with the desired playback speed. The interpolated or thinned audio data is transferred to the output buffer memory 8. The audio data stored in the output buffer memory 8 is read out by a clock signal having a constant period t, and is outputted to an output terminal 10 via a DA converter 9.

As described in detail, the present invention includes a frame dividing means (in the embodiment, consisting of an input buffer memory 3, etc.) that divides recorded audio data into frames of a fixed length, and a silent period for each frame. , a detection means for detecting whether it corresponds to a voiceless consonant section, a stationary vowel section, or an unsteady section (in the embodiment, it consists of a silence detection section 4 and a stationary determination section 5).
and a fundamental period extracting means (consisting of a fundamental period extracting section 6 in the embodiment) for extracting the fundamental period of speech in a stationary vowel interval, and a fundamental period extracting means (consisting of a fundamental period extracting section 6 in the embodiment), which extracts the fundamental period of speech in a stationary vowel interval in which the speech waveform changes periodically. Since the corresponding frame is played back after the audio data contained in the frame is interpolated or thinned out using the basic period of the audio as a unit, discontinuous parts can be reduced. In addition, for frames corresponding to unsteady sections where the audio waveform changes rapidly, the audio data contained in the frame is played back as is, so distortion of the reproduced sound can be suppressed, and therefore the main According to the invention, there is an advantage that high quality reproduced sound can be obtained even when the reproduction speed is changed. Therefore, the present invention can be applied to various audio processing applications such as high-speed playback tape recorders, high-speed playback of VTRs, high-speed reading of answering machines, voice message editing of voice mail systems, dictation machines using low-speed playback, language practice machines, etc. It is very effective when applied to equipment.

[Brief explanation of drawings]

FIG. 3 is an explanatory diagram of an interpolation method and a thinning method for audio data, respectively. 1 is an input terminal, 2 is an AD converter, 3 is an input buffer memory, 4 is a silence detection section, 5 is a stationarity determination section, 6 is a fundamental period detection section, 7 is an interpolation/decimation section, 8 is an output buffer memory, 9 is a DA converter, and 10 is an output terminal. Patent Applicant Nippon Telegraph and Telephone Public Corporation Representative Patent Attorney Hisa Gobu Tamamushi (3 others)'M 1
Figure-ITI- Figure 3

Claims (1)

[Claims] In an audio playback speed control method for controlling the playback speed of recorded audio data, there is provided a frame dividing means for dividing the audio data into frames of a constant length, and each frame is divided into a silent section, a silent consonant section, and a steady vowel section. , a detection means for detecting which of the non-stationary intervals corresponds to the voice, and a fundamental period extraction means for extracting the fundamental period of the voice in the steady vowel interval, which corresponds to the silent interval and the voiceless consonant interval. As for the frame, the audio data contained in the frame is interpolated or thinned out in units of a certain length, and then played back, and the frame corresponding to the stationary vowel section is played back. The audio data is interpolated or thinned out using the basic period extracted by the basic period extraction means as a unit, and then played back, and for frames corresponding to the non-stationary section, the audio data contained in the frame is played back as is. An audio playback speed control method characterized by: