JP5326796B2 - Playback device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a reproduction device creating a shortened voice signal with little sense of incongruity. <P>SOLUTION: A voice data is blocked for each segment to create voice blocks, and the voice signal of the voice block is discriminated to be a voiced sound, voiceless sound or silence, and a position where the voice starts is set to be a start point. When it is voiced sound, a deemed end point is determined on a time axis by using a presumed approximation line based on peak values of a plurality of voice waveforms which are close to an end point, and when it is voiceless sound, the end point is determined by level detection. Thus, a silence section included in the voiced sound, a voiceless section which can be shortened in terms of audibility and a silence section which are included in the voiceless sound, and a silence section where a voice data does not exist, are cut, and the voice signals are connected together again. <P>COPYRIGHT: (C)2011,JPO&amp;INPIT

Description

The present invention relates to a device for reproducing or recording / reproducing a sound composed of voiced voice, unvoiced sound and silent sound constituting a voice, and particularly relates to a function for listening to a reproduced voice in a short time or for clearly listening, Alternatively, the present invention relates to a capacity reduction of a storage device that stores an audio signal, and relates to a playback device that can minimize deterioration in sound quality without performing complicated calculations.

As a function installed in the playback device, various methods are used for speaking speed conversion technology for listening to the playback sound in a short time or for clearly listening. First, when the playback speed is simply increased with respect to the time axis, the frequency of the reproduced voice increases in proportion to the ratio of the increased speed, which makes it difficult to hear. In addition, when a portion below the detection level set by the input is simply skipped on the time axis and reproduced, it is not possible to skip an audio signal having no low level region. Therefore, in order to solve this problem, as disclosed in Patent Document 1, the pitch period of the voice input signal is extracted, and a weight window function is applied to the voice data for two pitch periods according to the pitch period. A time axis expansion and compression technique based on digital signal processing that performs time axis compression over time is employed.
Further, as described in Patent Document 2, the voice section and the silent section are identified, the time axis is compressed for the voice signal determined as voiced and silent, and the voice signal determined as unvoiced sound is recorded. The time axis is not compressed, or the time axis is compressed at a compression rate lower than the time axis compression ratio. Some are configured to identify a voice section and a silent section, delete the silent section, perform expansion / compression control by pitch synchronization of the voice section, or switch the thinning process according to the playback speed. .

JP-A-1-233835 JP-A-7-129198

  In such a conventional playback device, the problem to be solved is that when the voice segment and the silent segment are uniformly time-axis expanded and compressed, the clarity of the part to be heard deteriorates or the time axis expands. Audio data including compressed silence data is generated. Further, in the time-base expansion compression of voiced sound, when a sound signal of two pitches having periodicity is compressed to one pitch, a weight function W is multiplied by the front period P1, and an opposite weight is applied to the rear period P2. Since complicated signal processing is required to multiply the function 1-W and add each to one, the processing load is large and high-speed computation has to be executed.

  The present invention ensures a clear and uncomfortable feeling by leaving the minimum voiced sound and unvoiced sound for listening as speech without using a complicated signal processing method that requires high-speed computation. The present invention has been made for the purpose of providing a playback apparatus that produces a shortened audio signal with a small amount of noise.

This onset Ming is an apparatus for reproducing or recording and reproducing sound comprising a voiced unvoiced silence constituting the speech, the starting point for determining the starting point is detected that the audio signal data is present by the level detection Level detection circuit, a voice property identification circuit for identifying whether a block of voice data is voiced, unvoiced or silent, and a block of voice data identified by the voice property discrimination circuit for each segment to determine a voice clump An audio signal peak detection circuit that detects peak values of a plurality of waveforms close to the end of the audio waveform when the audio data block determined by the audio block identification circuit and the audio property identification circuit is voiced sound; The speech determined by the speech characterization discriminating circuit, and the speech qualification circuit, which estimates the terminal endpoint from the estimated approximate line formed by the envelopes of a plurality of peak values If the data block is an unvoiced sound, the end point level detection circuit for detecting the end of the voice data by level detection, and the start and end points determined by the deemed end point estimation circuit and the end point level detection circuit A voice segment cut circuit that cuts the data of the original voice signal based on the above information and a voice junction circuit that regenerates by connecting the cut voice data, and in the case of voiced sound, the deemed termination point Audio data in a period from the end point determined by the estimation circuit to the start point of the next block determined by the start point level detection circuit is cut, and in the case of unvoiced sound, the end point level detection circuit determines The voice data in the period from the end point to the start point of the next block determined by the start point level detection circuit is cut, and the silent section is cut all. This is a re-connected one, and it eliminates the gap between words and the silent part that occurs in one word and the silent part that can be shortened in the sense of hearing contained in the unvoiced sound. It has the effect that it can be created.
Furthermore, the present invention is provided with a storage circuit for storing the audio output output from the audio signal coupling circuit, and has the effect of reducing the storage capacity required for storage by cutting the audio data. Have.

In the case of voiced sound, the playback device of the present invention estimates a deemed end point from an estimated approximate line formed by an envelope of the peak value of the speech signal near the end of the speech block, and then estimates the following end point from the estimated assumed end point. Cut the audio data in the period up to the start point of the block of audio blocks, and in the case of unvoiced sound, cut the audio data in the period from the end point determined by level detection to the start point of the next block of audio blocks, The silence section is controlled so as to cut everything, and is connected again to generate a speech signal, shortening the gap between words and the silence contained in one word and the audibility contained in the silent sound. Since the voice signal from which the unvoiced portion that can be generated is deleted can be generated, there is an advantage that a shortened voice signal with less sense of incongruity can be created.
Further, by storing the produced audio signal in the storage circuit in the form of a shortened audio signal, an advantageous effect that the storage capacity necessary for storage can be reduced is obtained.

FIG. 3 is a block diagram of a signal processing portion of the playback apparatus according to the first embodiment. The figure which shows the audio | voice signal waveform by which the audio | voice identification of the reproducing | regenerating apparatus concerning Embodiment 1 was identified and divided into sections The figure which shows the audio | voice output waveform after the signal synthesis | combination of the reproducing | regenerating apparatus concerning Embodiment 1. FIG. Approximate line relationship diagram for estimating and calculating the end point of the playback apparatus according to the first embodiment only

    Hereinafter, the best mode for carrying out the reproducing apparatus of the present invention will be described in detail with reference to FIGS.

(Embodiment 1)
FIG. 1 is a block diagram of a signal processing portion of a playback apparatus according to the first embodiment of the present invention, and FIG. 2 is a voice signal waveform diagram in which voice is identified and divided into sections according to the first embodiment,
3 is an audio output waveform after signal synthesis in the first embodiment, and FIG. 4 is an approximate line relationship diagram for estimating and calculating a deemed end point in the first embodiment.
In FIG. 1, 1 is an audio input, 2 is a start point level detection circuit, 3 is an audio property identification circuit, 4 is an audio block circuit, 5 is a voiced signal path, 6 is an unvoiced signal path, and 7 is silent. , 8 is an audio signal peak detection circuit, 9 is a deemed termination point estimation circuit, 10 is a termination point level detection circuit, 11 is an audio segment cut circuit, 12 is an audio signal junction circuit, and 13 is an audio output. ing.

  The block configuration diagram of the signal processing portion in the first embodiment configured as described above will be described below with reference to FIGS. 2, 3 and 4. FIG.

First, a block diagram of the signal processing portion of FIG. 1 is obtained by extracting a signal processing portion that is a feature of the present invention from the entire playback apparatus. An example of the audio signal waveform on the time axis is the audio signal waveform diagram obtained by audio identification and divided into sections in FIG. An audio signal composed of an analog signal is input from the audio input 1. The start point level detection circuit 2 detects the level of the audio signal, and determines that an audio signal has been input when a level equal to or higher than a predetermined threshold is detected. When it is determined that there is an input, the start point is marked on this time axis. This corresponds to point A, point D, and point G in the audio signal waveform diagram shown in FIG. Depending on the threshold level set at this time, the level judged as silence will change, and if the threshold level is increased too much, there is a possibility that the head will be cut off at the beginning of the voice. Setting an appropriate level is an important factor. At this time, the start of the audio signal, that is, the start point can be determined, but it is not known whether the audio signal that follows is a voiced sound or an unvoiced sound. The audio signal to which the start end point is attached by the start end point level detection circuit 2 is sent to the sound property identification circuit 3. The voice property identification circuit 3 discriminates whether the voice signal is voiced, unvoiced or silent, but the processing becomes complicated if it is determined for each waveform of the voice signal composed of various frequency components. High signal processing capability is also required. Since the present invention is based on the premise of processing a gap between words and a silent portion generated in one word, the voice block circuit 4 blocks voice data for each segment and determines a voice clump. To do. There are various methods for blocking audio data, and a simple method is to divide the audio data based on the presence or absence of a signal level. Further, if the frequency component, the level, and the envelope shape of the audio signal are used to make the block with high accuracy, it is possible to prevent the head from being cut off.
The speech property identification circuit 3 determines whether a block of speech data is a voiced sound, an unvoiced sound, or a silent sound by judging the frequency components and the sound level of the block of the sound signal that has been blocked.
As in the classification by the voice blocking circuit shown in the upper part of FIG. 2, the sound is silent until point A, the voice is sound from point A to D, the voiceless sound is from point D to G, and the voice is sound from point G to K. First, classification is performed. In order to determine silence, silence can be recognized by determining whether there is an audio signal. In order to judge whether the sound is voiced or unvoiced, it is necessary to make a judgment based on the frequency and the sound level that make up the sound signal. A voiced sound is composed of a low-frequency waveform with a level fluctuation. With regard to the method for discriminating unvoiced sounds and voiced sounds, a great amount of technical data and patents have been published and are not the main purpose of the present invention, so they are omitted here.

  The voice data classified into three types of voiced sound, unvoiced sound, and silent sound by the voice property identification circuit 3 proceeds to different processes. In the case of voiced sound, a speech signal peak detection circuit 8 that detects peak values of a plurality of waveforms close to the end of the speech waveform, and a deemed termination that estimates a deemed termination point from an estimated approximate line formed by an envelope of the plurality of peak values The point estimation circuit 9 divides into the following three. The first is a section from point A to B, which is identified as voiced sound in the blocking circuit but has the nature of unvoiced sound, and the second is from point B which is completely classified as voiced sound. C, segment from point H to J, and the third segmented into point C marked as a deemed end point and silent part consisting of segment from point C to D and point J to K delimited by point J Is done. Here, since the method of determining the deemed end point is the gist of the present invention, it will be described in detail later. In the case of an unvoiced sound, the end point level detection circuit 10 detects the end of the voice data by level detection and marks the end point F. Thereby, it can be subdivided into a section from point D to E which is an unvoiced sound that can be determined as unnecessary included in the unvoiced sound, a section from point E to F that can be completely classified as unvoiced sound, and a section from F to G that can be regarded as silent.

  The difference between the subdivision of the voiced sound block and the subdivision of the unvoiced sound block is due to the characteristics of the sound signal. Unvoiced sounds are composed of continuous waveforms with low levels and high frequencies. The end of this waveform converges to silence while gradually decreasing the level. The determination of the end point is ambiguous, and it is necessary to determine the end point F by level detection using a threshold value. In addition, since it converges gently at a low level, even if the data is cut on the time axis immediately after the sound disappears and connected to the head of the next voice block, it is possible to create a relatively uncomfortable voice. Because.

The voice section cut circuit 11 performs an operation of cutting voice signal data based on the information of the start and end points determined by the deemed end point estimation circuit 9 and the end point level detection circuit 10. The blocks after detailed classification are voiced sounds, unvoiced parts included in voiced sounds, unvoiced parts included in voiced sounds, unvoiced sounds, unvoiced parts included in unvoiced sounds, and completely silent. . Here, the section from point C to D which is a silent part included in the voiced sound, the section from point J to K, the point D to E which is an unnecessary silent part which does not have a significant effect on the sound quality even if it is included in the unvoiced sound. The voice data of the section from point F to G, which is a silent part included in the silent sound, is cut. In response to the cut audio signal data, the audio signal joining circuit 12 directly connects the position where the wave height level determined as the end point is 0 and the position where the wave height level determined as the start point is 0, to the audio data. Join. FIG. 3 is an example of an audio output obtained by cutting and connecting the audio signal waveform diagram of FIG.
In this first embodiment, the unvoiced sound part included in the voiced sound and the slight silence part between the voiced sounds are not cut, but are detected and cut to produce a further shortened sound signal. Is also possible. In this case, the section from point M to B and the section from point N to H shown in FIG. 2 are cut.
Here, with reference to FIG. 4, a method for determining the end point only for the voiced sound described above will be described. In the case of unvoiced sound, the sound converges from the level attenuation at a low level, but in the case of voiced sound, the sound is significantly different from that, and most attenuation of standard voiced sound has the following characteristics. ing. First, the angle of the intersection of the estimated approximate line of the peak values at which attenuation begins and the time axis intersect with α, and the extended line of the approximate line created by the multiple peak values approaching the terminal side as attenuation further proceeds If the angle of the intersection with the time axis is β, β> α. That is, the audio signal does not remain after the intersection of the time axis and the extension line of the estimated approximate line where the attenuation starts before the estimated approximate line in the second stage. If this intersection is regarded as a termination point and used as a point for cutting the previous audio signal, it is possible to reproduce a synthesized sound without any sense of incongruity. I don't know why this point is good, but after processing many audio signals and repeatedly listening to them, the effect was confirmed. In addition, when the waveforms of a plurality of audio signals were actually confirmed and the peak value forming the estimated approximate line was 3 to 5 points, it was data for which a good end point was easily obtained.
Further, due to the nature of an audio signal, the frequency components and peak values constituting it are not linear but have waved components. Therefore, when the peak point of the waveform is strictly pursued and an approximate line is drawn, it is considered that the assumed end point to be obtained cannot be found. Thus, a determination method including the ambiguity is required for the peak value when calculating the deemed end point.
In the description of the first embodiment, an estimated approximate line by a figure is used. However, as a specific approximation creation method, digital processing is performed by using creation of an approximate line by primary linear interpolation. You can also find the end point with.

  In the case of voiced sound, the playback device of the present invention cuts audio data for a period from the end point determined by the deemed end point estimation circuit to the start point of the next block determined by the start point level detection circuit. In the case of unvoiced sound, the audio data in the period from the end point determined by the end point level detection circuit to the start point of the next block determined by the start point level detection circuit is cut, and the silent section is By controlling everything to cut and reconnecting them, the gap between words and the silent part generated in one word and the unvoiced part included in the unvoiced sound can be deleted to reduce the sense of incongruity. In a device that can be used for quick listening as speech speed conversion and that records the voice data, the storage capacity required for storage can be reduced. It can also be applied to applications.

DESCRIPTION OF SYMBOLS 1 Voice input 2 Start point level detection circuit 3 Voice property identification circuit 4 Voice blocking circuit 5 Voiced signal path 6 Unvoiced signal path 7 Silent signal path 8 Voice signal peak detection circuit 9 Deemed termination point estimation circuit 10 Termination Point level detection circuit 11 Voice segment cut circuit 12 Audio signal joining circuit 13 Audio output

Claims (3)

A device for reproducing or recording / reproducing a sound composed of voiced sound, unvoiced sound and silent sound that constitutes a sound,
A level detection circuit for a starting point that detects the presence of audio signal data by level detection and determines a starting point;
A voice property identification circuit for identifying whether one block of voice data is voiced, unvoiced or silent;
A voice blocking circuit that blocks voice data identified by the voice property identification circuit for each segment and determines a voice chunk;
When the voice data block determined by the voice property identification circuit is voiced sound, a voice signal peak detection circuit that detects peak values of a plurality of waveforms near the end of the voice waveform;
A deemed end point estimation circuit for estimating a deemed end point from an estimated approximate line formed by an envelope of the plurality of peak values;
When the voice data block determined by the voice property identification circuit is an unvoiced sound, a termination point level detection circuit that detects the end of the voice data by level detection;
A starting point determined by the level detecting circuit for the starting point, the considered termination point estimating circuit and the voice interval cutting circuit for cutting the data of the audio signal based on information of the final end point that is determined by the level detecting circuit terminating point When,
It is equipped with an audio junction circuit that regenerates by connecting the audio data after cutting,
In the case of voiced sound, the speech data of the period from the start point determined by the start point level detection circuit to the end point determined by the deemed end point estimation circuit, and determined by the deemed end point estimation circuit The voice data in the period from the next end point to the start point of the next block determined by the start point level detection circuit is cut, and in the case of unvoiced sound, the start point determined by the start point level detection circuit To the end point determined by the end point level detection circuit and the next end point determined by the start point level detection circuit from the next end point determined by the end point level detection circuit. of cutting the audio data time to the starting point of the mass, Ru joined again performs control silent section to cut all playback devices. A storage circuit for storing the audio output output from the audio signal combining circuit;
2. The reproducing apparatus according to claim 1, wherein the storage capacity necessary for storing the audio data is reduced by cutting the audio data. 2. The reproducing apparatus according to claim 1, wherein the deemed terminal point estimation circuit obtains a deemed terminal point using an approximate line estimated by linear linear interpolation.

Images