JP3249567B2 - Method and apparatus for converting speech speed - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech speed conversion method and apparatus suitable for listening to a voice of a hearing impaired person or an elderly person.

[0002]

SUMMARY OF THE INVENTION The present invention relates to a speech speed conversion method and apparatus suitable for hearing voices of hearing-impaired persons, elderly people, and the like. By shortening the silent section between sentences to the minimum as long as there is no unpleasant sensation, and changing the speech speed based on a certain rule, the speech time is maintained slowly while maintaining the speech time of the original voice as a whole. It is intended to convert the sound into good sound which is easy to hear.

[0003]

2. Description of the Related Art A technology for converting speech speed while maintaining quality is in the process of development, and a technology that takes into account a "deviation" from real time (frame) has not been developed.

[0004]

By uniformly lowering only the speech speed of speech, it is possible to make it much easier to hear, especially for the elderly and the hearing impaired.
This operation inevitably extends the speech utterance time.
However, in broadcasting and reading cassettes, the utterance time of the voice before expansion is uttered so as to fit within the predetermined time.Therefore, if such a voice is expanded, it may not be able to fit within the above-mentioned time limit. Occurs. Also, in the case where audio and video are provided in synchronization, such as on a television, if only audio is expanded, a time lag may occur between the video and the video, which may adversely affect listening. Conceivable.

[0005] An object of the present invention is to solve the above-mentioned problem associated with the temporal "shift" by appropriately reducing the speech speed of a meaningful portion in the uttered voice, and reversely reducing the other portions. Accordingly, it is an object of the present invention to provide a speech speed conversion method and apparatus for converting speech into a slow and easy-to-hear sound as a whole without substantially extending the speech time by speeding up.

[0006]

In order to achieve the above object, according to the first aspect of the present invention, there is provided a method for converting a speech speed, which comprises the steps of: According to a change in the pitch (fundamental frequency), the speech speed is reduced at a high pitch and the speech speed is increased at a low pitch.

In order to achieve the above object, a speech speed conversion method according to a second aspect of the present invention provides a speech speed conversion method in which, when the speech speed is reduced, a section of a vocalization and a next vocalization are slowly set at a start point of this section. Is set, and the speech speed is gradually increased toward the end point by following a rough change in the fundamental frequency of the voice.

[0008] Here, when the speech speed is reduced, it is further characterized in that the silent section between sentences is shortened as short as possible within a range in which the sense of incongruity obtained by an experiment is obtained in advance. Further, the time as short as possible within a range in which there is no uncomfortable feeling in the hearing, which is previously determined by an experiment, is in a range from 862 ms to almost 1000 ms.

In order to achieve the above object, a speech speed conversion apparatus according to a fifth aspect of the present invention is a speech speed conversion apparatus, comprising: a voice identification unit for identifying a voice signal as voiced, unvoiced, and silent; and a silent section identified by the voice identification unit. A silent section determining means for determining whether or not there is a silent section between sentences; and, if the silent section determining means determines that there is a silent section between sentences, the silent section is determined in advance by an experiment in a range where there is no auditory discomfort. Silent section shortening means for shortening to a time as short as possible, voiced section determining means for determining whether or not the voiced section identified by the voice identifying means is the start of voiced voice, and start of voiced voice by the voiced section determining means. If it is determined, set a slow speech speed at the start point of this section in units of the vocal and the next vocal section, and follow the rough change of the fundamental frequency of the voice toward the end point. hand S and characterized by including a voiced segment stretching means for performing speech speed conversion processing to increase the speech rate on.

[0010]

According to the present invention, when the speed at which a received voice is generated (speaking speed) is reduced, the voice speed is reduced at a high pitch and the voice speed is reduced at a low pitch in accordance with a change in the pitch (fundamental frequency) of the voice. The feature is to speed up the talk. Also, the present invention
When decreasing the speech speed, set a slow speech speed at the start point of this section in units of the voice and the next voice, and roughly change the fundamental frequency of the voice toward the end point The feature is that the speech speed is gradually increased to follow.

Further, in the present invention, attention is paid to a silent section between sentences, and the silent section between sentences is shortened to a time as short as possible within a range in which the sense of incongruity previously obtained by an experiment is not affected. As an example, a time as short as possible within a range where there is no uncomfortable feeling is 862 ms to almost 1000 times.
ms.

Therefore, according to the present invention, it is possible to listen to a slow and easy-to-listen sound desired by the listener within the real time frame without extending the utterance time.

[0013]

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

(1) Apparatus Configuration FIG. 1 shows an apparatus configuration according to an embodiment of the present invention. The audio input circuit 1 is a circuit having a general configuration for inputting an audio signal. If necessary, for example, a microphone, a tone circuit, an analog / digital converter, an audio storage / reproduction (recording) circuit,
Audio storage media (for example, IC memory, hard disk,
Floppy disk or VTR), and an interface circuit. The CPU (Central Processing Unit) 2 controls the entire apparatus, performs calculations, and the like. For example, a known one-chip microcomputer or personal computer can be applied. Program memory (PRO
M) 3 stores in advance control procedures (programs), tables, constants, and the like as shown in FIG.

The input buffer 4 and the processing buffer 5 are C
The PU 2 is secured in a RAM (random access memory) (not shown) used as a work area, and digital audio signals input from the audio input circuit 1 are temporarily stored in the input buffer 4 sequentially in frame units described later. The audio signal stored in the input buffer 4 is temporarily stored in the processing buffer 5 for each segment described later. The file 6 is a memory for storing a voice signal which has been subjected to processing for expanding a voiced section and shortening a silent section according to the present invention.
In addition to the AM, an audio storage medium such as an IC memory or a floppy disk can be applied.

The audio output circuit 7 is a circuit having a general configuration for outputting the audio signal in the file 6 to the outside. If necessary, for example, an interface circuit, a digital / analog converter, a speaker, a recording device (or a broadcast device) Equipment). Of course, it is also possible to harden all the procedures shown in FIG.

(2) Operation Example FIG. 2 shows an operation procedure of an embodiment of the present invention. In the present embodiment, when the speed of uttering the listening sound (speaking speed) is reduced,
The silent section is shortened to the shortest possible time without a sense of incongruity, and a significant portion in the uttered voice is a place where the pitch (fundamental frequency) of the voice is usually high, and the place where the pitch is high is at the start of vocalization. Focusing on this, set a slow speech speed at the start point of this section in units of the vocal and the next vocal section, and follow the rough change of the fundamental frequency of the voice toward the end point And then gradually increase the speaking speed.

Step S1: First, the voice input circuit 1
The input audio signal is cut out into a fixed-length portion called a frame and stored in the input buffer 4. In this embodiment, the frame length is, for example, 3.3 ms.

Step S2: Voiced, unvoiced,
Performs silence determination. As the determination method, for example, a known autocorrelation method and a zero-cross method can be applied. Of course, other determination methods may be used. In principle, input sounds other than voiced and unvoiced voices (for example, low-level noise and background sounds) uttered by humans are processed as silent.

Step S3: If the type of the current and previous frames is the same, return to step S1, and if different, for example, change from voiced to unvoiced, proceed to the subsequent stage. As a result, the same type (section) of voice is stored in the input buffer 4.

Step S4: From the average of the number of mora uttered in one second, the threshold values Th1 and Th described later are calculated.
2. Set Th3. Mora corresponds to the length of one syllable including a short vowel. In Japanese, it is almost equivalent to one character of kana (two characters in MUON). The processing in step S4 may be performed only at the initial stage or at predetermined intervals.

Step S5: A section starting from unvoiced or unvoiced and ending with voiced is one block (B _n : n = 1, 2, 2)
…). In this block, it is roughly classified into a silent section (a _n ), an unvoiced section (b _n ), and a voiced section (C _n ) according to the determination in step S2, and each section is sent to the following processing system. Can be The time at the boundary between b ₁ and c ₁ is expressed as t _{1, s,} and the first voice is α1 (see FIG. 3).

Step S6: As shown in FIG. 3, the time between the start point (t _{n, s} ) of the _nth voiced section and the end point (t _{n-1, e} ) of one previous voiced section. Interval T _n (T _n = t
_{n, s} -t _{n-1, e} ).

[0024] Step S7: comparing the threshold value Th1 for determining T _n and voice stand. If T _n exceeds a certain threshold value Th1, t _{n, s}
Is determined to be the voice α _m (see FIG. 3), and step S
Proceed to 8. If there is no previous voiced section at the start of this processing, the process jumps to step S11 described later.

Step S8: The previous voice α _m-1 and 1
The range of the end point t _{n-1, e} of the preceding voiced section is defined as one segment. In the example of FIG. 3, T ₅ = t _{6, s} −t _{5, e} > Th1
Then _, the time point of t6 _{, s} is voiced α ₂ , and the section (t _{5, e} −
t _{1, s} ) becomes one segment. Then, step S1
The expansion ratio r _s of the voiced section length of the start point of one segment stored so far in the processing buffer 5 is set to a predetermined value within the range of 1 ≦ r _s ≦ 2 by the processing of 1, S12, and S15. And stretch. The stretching magnification is gradually decreased toward the end point of this segment, stretching magnification r _e voiced section length of the end point is made to be 0.7 ≦ r _e ≦ 1. FIG. 4 shows an example of a method of obtaining the expansion factor of the voiced section belonging to the segment 1 in FIG. Voiced c ₁ of the segment starting point is stretched _{c 1 '= r s · c} 1, c 2 is c
₂ ′ = r ₂ · c ₂ Voiced c ₅ segments end point becomes a _{c 5 '= r e · c} 5, r e is because a r _e ≦ 1, in practice is shortened. No processing is performed on the silent section a _n and the unvoiced section b _n other than the voiced section, and they are unchanged.

That is, in general, the voice of the vocal part (the first half of one unit) is often important in terms of meaning, so that the audibility is improved by appropriately reducing the speech speed as described above. I do. The change in speech speed is determined by the appropriate function f
It is changed using (t). In the present embodiment, a cosine function as shown in FIG. 4 is used as an example. In this case, f
(T) is expressed by the following equation (1).

[0027]

(Equation 1)

Step S9: The voice data whose speech speed has been converted in step S8 is dropped to the file 6.

Step S10: The processing buffer 5 is cleared.

[0030] Step S11: In the step S7 T _n ≦ T
In the case of h1, or when step S10 is processed, the process proceeds to step S11. If a negative determination is made in step S7, it is determined that the voiced section falls within one unit, and this voiced section is stored in the processing buffer 5. If step S10 has been reached, the voiced section at the start of the vocalization is stored in the processing buffer 5
Will be stored. The input buffer 4 is cleared for the processing of the next audio data, and if an instruction to end this processing operation has not been issued (step S16), the process returns to step S1.

Step S12: The unvoiced section is always transferred from the input buffer 4 to the processing buffer 5 and stored. After that, the input buffer 4 is cleared, and step S1
After step 6, the process returns to step S1.

Step S13: If the section for each type of voice is a silent section, the length of the silent section is compared with a threshold value Th2 for determining a break (punctuation) between sentences. When the silent section exceeds Th2, the silent section is determined to be a break (punctuation) between sentences, and the process proceeds to the next step S14. Otherwise, the process jumps to step S15.

Step S14: The silent section determined as a period is shortened by the following procedure.

[0034] In order to shorten to the shortest without any discomfort in the sense of hearing,
The time length of the shortened silent section is the threshold value Th3. If the time length a _n of the silent section, the time length d _n of the section to be deleted, the time length of the silent section after the deletion was e _n, e _n
, As shown in (B) of FIG. 5, a _{e n = a n -d n ···} (2). In this case, the specified error silence range during analysis, because there is a possibility of identifying as part of a long silence even unvoiced portion, from the beginning of a _n, instead of deleting the d _n, 5 as shown in the (a), d _n from the center point of a _n
Delete part. Further, both ends of the d _n, smoothes over the taper of several ms, to prevent the occurrence of this the click sound. Since the silence here includes a sound other than the voice generated from a person as described above, this smoothing process is useful.

[0035] In the above equation (2) e _n value e _n ≧ Th
It may be set as a variable value in the range of 3, but a fixed value close to Th3 to e _n order to simplify the processing (eg, 862
If set to ms), d _n from the above equation (2) is a variable value that varies by a _n. Next, the process proceeds to step S15.

Step S15: The silent section is stored in the processing buffer 5. Clear the input buffer 4 and execute step S1
After step 6, the process returns to step S1.

Step S16: If there is no sound signal data in the sound input circuit 1, or if there is a work stop command, the present processing routine ends and returns to the main standby routine and the like.

(3) Experimental Example In the experimental example of this embodiment, a news sentence of 136 seconds was applied. In this case, the average of the speech speed was 9.6 mora / second. Th2 = Th3 = Th1 = 3
50 ms, and Th2 = Th3 = 1000 ms.
At this time, according to a psychological experiment, as for the speech speed control, the speech speed at the start point in one unit (the expansion rate of the voiced section length) is 1.0 to 1.3 times that of the original voice, and the speech speed at the end point is 0. naturalness in the range of .9～1.0 times, high evaluation is obtained in clarity, also, for the shortening of the silent section, if shortened silence section (e _n) is them 862ms present at a minimum, the audibility discomfort It was found that there was not.

From the results, the speech speed was changed from a slow speech speed of 1.2 times to a fast speech speed of 0.92 times, and a long silent section ("ma" between sentences) was shortened to 1200 ms. It was confirmed that the speech times of the original voice and the converted voice coincided with each other, and that a good voice speed converted voice could be obtained.

(4) Other Embodiments During the processing in step S8 (see FIG. 2) of the above-described embodiment, high-quality sound quality can be maintained by performing processing so that the pitch does not change even when the speech speed changes. . As this processing method, for example, a method of processing an audio signal disclosed in Japanese Patent Application No. 3-245960 “Speech Rate Control Type Hearing Aid Method and Apparatus” is suitable.

Further, stretching magnification r _s voiced interval length in the above embodiment, although a constant value the time length e _n, etc. are previously decided after deletion of r _e silent section, the user desires from the dial or a keyboard May be a variable value that can be set to the value of. This makes it easier to perform, for example, editing work that matches the wishes of the viewer or that fits exactly within the broadcast time.

Also, instead of the voiced section expansion processing in the above embodiment, the pitch (fundamental frequency) of the voice is directly detected by a known pitch extraction method, and the speech speed is changed at a high pitch according to the change in pitch. May be relaxed, and the speech speed may be increased in low places.

[0043]

As described above, according to the present invention,
When decreasing the speed at which the listening sound is generated (speaking speed), the speaking speed is slowed at a high pitch and the spoken speed is increased at a low pitch in accordance with a change in the pitch (fundamental frequency) of the sound. In addition, set a slow speech speed at the start point of this section in units of the vocal section and the next vocal section, and gradually follow the rough change of the fundamental frequency of the voice toward the end point. Since the speech speed is increased, and the silent section between sentences is shortened to the shortest possible time within the range of hearing perceived incongruity determined in advance by experiments,
An effect is obtained in which the speech can be converted into a good voice that is easy to listen to slowly as a whole while keeping the speech time at the speech time of the original voice.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an apparatus configuration according to an embodiment of the present invention.

FIG. 2 is a flowchart showing processing contents of one embodiment of the present invention.

FIG. 3 is a diagram showing a segmentation of audio data based on a process according to an embodiment of the present invention.

FIG. 4 is a timing chart showing a change in speech speed in one embodiment of the present invention.

FIG. 5 is an original waveform (A) based on the processing of one embodiment of the present invention.
FIG. 8 is a waveform diagram showing a waveform (B) obtained by shortening a long silent section between sentences.

[Explanation of symbols]

Reference Signs List 1 voice input circuit 2 CPU 3 PROM 4 input buffer 5 processing buffer 6 file 7 voice output circuit a _n silent section b _n unvoiced section c _n voiced section c _n ′ expanded voiced section B _n voiced beginning with voiceless and ending with voiceless Section Th1 Threshold value for discriminating voice-throat Th2 Threshold value for discriminating breaks (phrases) between sentences r _s Extension rate of voiced section length at start point r _e Length of voiced section length at end point Stretch magnification

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-1-93795 (JP, A) IEICE Technical Report SP 92-56 A method for absorbing time extension accompanying speech speed conversion (58 ) Surveyed field (Int.Cl. ⁷ , DB name) G10L 21/04

Claims (5)

(57) [Claims] When the speed at which a received voice is generated (hereinafter referred to as a voice speed) is reduced, the voice speed is reduced at a high pitch and the voice speed is lowered at a low pitch in accordance with a change in voice pitch (fundamental frequency). Is a speech speed conversion method characterized by increasing the speech speed. 2. When the speech speed is reduced, a slow speech speed is set at the start point of a section of a vocal utterance and the next vocal utterance unit, and the basic speech is set toward the end point. A speech speed conversion method characterized by gradually increasing the speech speed following a rough change in frequency. 3. The method according to claim 1, wherein, when the speech speed is reduced, a silent section between sentences is shortened to a time as short as possible within a range in which a sense of incongruity obtained by an experiment is obtained in advance.
Or the speech speed conversion method according to 2. 4. The time as short as possible within a range in which there is no uncomfortable feeling on hearing obtained in the experiment in advance is from 862 ms to almost 10
4. The speech speed conversion method according to claim 3, wherein the range is 00 ms. 5. Speech identification means for distinguishing a speech signal into voiced, unvoiced, and silence; silence section determination means for judging whether or not the silence section identified by the speech identification means is a silence section between sentences; When the silent section determining unit determines that the silent section between sentences is a silent section between sentences, a silent section shortening unit that shortens the silent section to a time as short as possible within a range where there is no uncomfortable feeling obtained by an experiment in advance; Voiced section determining means for determining whether or not the identified voiced section is the start of a voiced voice; and, if the voiced voiced section determining means determines that the voiced voice is to be started, the voiced voice and the next voiced voice are used as a unit. Voiced section decompression means that performs a speech rate conversion process that sets a slow speech rate at the start of the leverage section and gradually increases the speech rate following a rough change in the fundamental frequency of speech toward the end point. Speech speed converting device being characterized in that comprises a.