JPS592033B2 - Speech analysis and synthesis device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech analysis and synthesis device, and more particularly to a speech analysis and synthesis device that can be used, for example, in a digital speech transmission device.

For example, in order to reduce transmission capacity and efficiently transmit audio information, various low-speed digital audio transmission systems have recently been proposed.

If such a digital voice transmission system is used, for example, in a telephone line, the transmission capacity can be significantly reduced compared to conventional PCM, which is very advantageous for the use of the line. FIG. 1 is a schematic block diagram showing an example of a conventional speech analysis device which is the background of the present invention.

In this conventional example, a voice input from, for example, a microphone is provided to a formant extraction/sound source extraction circuit 11 by an input source 11a.

This extraction circuit 11 extracts only perceptually necessary information from the audio waveform. For example, formant strength and its frequency are extracted as spectral distribution information from the spectral distribution shown in FIG. Normally, from the strength or frequency of such formants, information of about three formants, the first formant, the second formant, and the third formant, as shown in FIG. 2, is extracted. Further, the sound source information includes information for identifying whether the voice input is voiced or unvoiced, pitch period, sound intensity, and the like. And this extraction circuit 11
These information extracted by A/
The signal is applied to the D converter 12. This A/D converter 12 converts the extracted information into digital information, and converts the extracted information into digital information.
It is applied to the speech synthesis section shown in the figure. The speech synthesis section is provided with an oscillator 21 whose resonant frequency is the pitch period.

The signal from this oscillator 21 is passed through an amplitude control circuit 22, such as an amplifier, as a sound source, and then to resonators 23, 24.
and 25. The resonant frequencies of the resonators 23, 24, and 25 are selected so that they resonate at the frequencies of the extracted first, second, and third formants, respectively. Then, the noise generator 26 for the audio state generates white noise, and the amplitude control circuit 2
Give to 7. The amplitude control circuit 27 controls the amplitude according to the strength. The voiced or unvoiced identification information extracted by the voice analysis section is given to the switch 28,
When the sound is a voiced sound, this switch 28 is set to the state shown in FIG. 3, and when the sound is an unvoiced sound, this switch 28 is switched. A signal from this switch 28 drives a speaker 29 via an amplifier (not shown) or the like. As described above, the conventional method for transmitting spectrum information has the advantage that the number of occupied bits of the line can be reduced. However, this conventional method has the following problems. That is, the spectral distribution fluctuates minutely as shown in FIG. 2, making it difficult to accurately identify peaks. Therefore, the quality of the synthesized sound cannot necessarily be said to be good. Therefore, the main object of the present invention is to provide a novel type of speech analysis and synthesis device.

Another object of the present invention is to provide a voice analysis and synthesis device that is simpler in construction and therefore less expensive. Still another object of the present invention is to improve the quality of synthesized speech. An object of the present invention is to provide a speech analysis and synthesis device. In summary, this invention is based on the principle of spectral preservation, which states that if the presence or absence of periodicity, fundamental period, and spectral distribution of speech are preserved, the original speech can be correctly restored. The data is analyzed using a bandpass filter and converted into digital data, which is then converted into analog data in the synthesis section to reproduce the spectral distribution and synthesize audio.

The stated objects and other objects and features of the invention will become more apparent from the following detailed description with reference to the drawings.

FIG. 4 is a schematic block diagram showing the speech analysis section 4 according to an embodiment of the present invention.

In the configuration, input audio is provided to input line 4a. Then, this audio input is given to, for example, n band pass filters 411, . bandpass filter 411,
...41n is, for example, 200 to 5000
They are used to pass frequency components of different bands in the audio band such as Hz, and are provided relatively evenly distributed in the band. Alternatively, the frequencies of these bandpass filters 411, . . . 41n may be selected so as to be concentrated in a specific frequency band. And the high pass filter 41n+1 is
This is for extracting unvoiced, ie, noise components of the audio input. These filters 411,...41n
The outputs of 41n+1 and 41n+1 are given to corresponding A/D converters 421, . . . 42n and 42n+, respectively. A/D converter 421 to 42n, 42n+1
are digital data 1 to N, . Derive n+1.

Note that these A/D converters 421 to 42n, 42n+
1 may of course be configured to include a kind of peak holding circuit such as a low pass filter.
Furthermore, the voiced/unvoiced detection circuit 43 is for detecting the voiced state or unvoiced state of the audio input, and detects whether the sound is voiced or unvoiced based on the magnitude, energy, etc. of the autocorrelation function of the audio waveform. .

Another voiced or unvoiced information from this detection circuit 43 is
It is derived as voiced/unvoiced data via the A/D converter 44. FIG. 5 is a schematic block diagram showing the speech synthesis section 5 of one embodiment of the present invention.

In the configuration, an A/D converter 4 included in the voice analysis section 4
Digital data from 21 to 42n, 42n+1 and 44 is given to this speech synthesis section 5. Then, data 1 to N. n+1 are latched into corresponding latch circuits 511 to 51n and 51n+1, respectively. These data 1 to N. n+1 represents each frequency band or the level of the noise component, and is converted into an analog voltage by the D/A converters 521 to 52n, 52n. On the other hand, sine wave oscillators 531 to 53n having an oscillation frequency equal to the center frequency of bandpass filters 411 to 41n included in the analysis section 4 are provided.

The outputs of these sine wave oscillators 531 to 53n are given as carrier waves to AM modulators 551 to 55n.
The AM modulators 551 to 55n use the analog signals from the corresponding D/A converters 521 to 52n as modulation signals, and apply amplitude modulation to the input sine wave signals. Also,
A noise generator 54 is provided to generate white noise for unvoiced sounds.

The noise component from this noise generator 54 is given to a gate circuit 56. This gate circuit 56 transmits noise components to the AM modulator 55 when there is no voice based on the voiced/unvoiced data from the A/D converter 44 included in the voice analysis section 4.
Give it to n+1. That is, the circuit for this noise only needs to be activated for unvoiced sounds. Therefore, without providing such a gate circuit 56, a noise component is constantly supplied to the AM modulator 55n+, and the AM modulator 55n+1 is enabled or activated only when there is an unvoiced sound. good. Then, the AM modulator 55n7 uses the noise component as a carrier wave and applies AM modulation to this noise component using the analog signal from the D/A converter 52n+1 as a modulation signal. Knob The outputs of the above-mentioned AM modulators 551 to 55n are given to the adder 57 when the sound is voiced, and the outputs of the AM modulators 551 to 55n are applied to the adder 57,
The output of is given to the adder 57 when it is an unvoiced sound.

Then, the adder 57 combines the signals given at that time and supplies the synthesized sum to the amplifier 58. The amplifier 58 is the speaker 59
As a result, the synthesized sound is generated from the speaker 59. Note that the level changes for each frequency component extracted by the bandpass filter are very gradual (approximately 50 Hz at most), so the sound pressure level data 1 to N, . n+1 is number 10
It is sufficient to give it every msec.

In addition, it is known that the phase characteristics are of little importance in perceiving speech, and that a phase shift of about 50 msec is not a problem, so this embodiment also does not provide a means to specifically specify the phase. do not have. FIG. 6 is a main circuit diagram showing a preferred embodiment of the speech synthesis section of FIG. 5. In particular, in FIG. 6, the sine wave oscillator 531. Although AM modulator 531 and adder 57 are shown in detail, it should be noted in advance that other corresponding components may be similarly configured. Sine wave oscillator 531 includes amplifier A1.

One input of the amplifier A1 is connected to a connection point between resistors R1 and R2, and the other input is connected to a connection point between a variable resistor R3 and a capacitor C. The other end of the resistor R1 is grounded, and the other end of the resistor R2 is connected to the output of the amplifier A1. Further, the other end of the variable resistor R3 is grounded, and the other end of the variable resistor C is connected to the connection point between the resistor R4 and the thermistor TH. The other end of resistor R4 is amplifier A1
The other end of the thermistor TH is grounded. In such an oscillation circuit, variable resistor R3 and capacitor C determine its oscillation frequency, and resistors R1 and R2 determine its output amplitude. In addition, the thermistor T
H is for keeping the output amplitude constant. That is, when the amplitude of the oscillation output becomes large, the thermistor TH changes its resistance and functions to suppress the amplitude. The oscillation output from such a sine wave oscillator 531 is given to the AM modulator 551 via a capacitor. AM modulator 551 includes an amplifier A2 that receives the output from oscillator 531 via resistor R11 at one input thereof. The other input of this amplifier A2 is grounded. A parallel circuit including a resistor Rl2 and a light receiving element CdS is connected between one input and the output of the amplifier A2. This light receiving element CdS constitutes a photocoupler PC in cooperation with the light emitting element LED. Then, the analog voltage from the D/A converter 521 is rectified and smoothed by the diode D1 and the capacitor C1 and then applied to the light emitting element LED. Therefore, the light emitting intensity of this light emitting element LED changes depending on the output, ie, data, from the D/A converter 521. On the other hand, the light receiving element CdS is such that the more light it receives, the smaller its resistance value becomes. And the gain of amplifier A2 is given by.

Therefore, the gain of this amplifier A2 changes depending on the brightness of the light emitting element LED. That is, the sine wave signal from the sine wave oscillator 531 input to the amplifier A2 is subjected to AM modulation by the output of the D/A converter 521. Then, the adder 57 receives the output from the similar AM modulator, and sends the output to the amplifier 5 via the amplifier A3.
8 gives the composite output.

It goes without saying that these sine wave oscillators, Ay5 modulators, etc. can be used with various circuit configurations in addition to the circuit configuration shown in FIG.

As described above, according to the present invention, input speech is analyzed by determining the levels of a plurality of frequency components of the speech input using several bandpass filters, and the synthesis section analyzes the bandpass filters by determining the levels of a plurality of frequency components of the speech input. Since multiple sine wave signals corresponding to the filters are AM-modulated according to their corresponding levels, and these AM-modulated signals are synthesized, it can be configured with a very simple circuit. .

Therefore, it is cheaper than the conventional one. Furthermore, according to the present invention, the compression rate of audio data is relatively high;
The clarity of synthesized speech is good.

[Brief explanation of drawings]

Fig. 1 is a schematic block diagram showing an example of a conventional speech analysis device, Fig. 2 is a waveform diagram showing an example of spectral distribution, and Fig. 3 is a diagram of a conventional speech synthesis device corresponding to Fig. 1. It is a schematic block diagram which shows an example. FIG. 4 is a schematic block diagram showing the speech analysis section 4 according to an embodiment of the present invention. FIG. 5 is a schematic block diagram showing the speech synthesis section 5 according to the first embodiment of the present invention. FIG. 6 is a circuit diagram showing essential parts of a preferred embodiment of the embodiment shown in FIG.
In the figure, 4 is a voice analysis section, 411 to 41n are band pass filters, 41n+1 is a high pass filter, 4
3 is a voiced/unvoiced detection circuit, 421 to 42n, 42n
+1 and 44 are A/D converters, 5 is a voice synthesis unit, 52
1 to 52n, 52n+1 are D/A converters, 531 to 53n are sine wave oscillators, 54 is a noise generator, 55
1 to 55n, 55.

Claims (1)

[Claims] 1. Audio input means for providing audio input; a plurality of bands selected to receive audio input from the audio input means and selectively pass different bands of the audio input; a pass filter, a plurality of level data generating means for receiving the output of a corresponding one of the plurality of band pass filters and generating data according to the output level thereof, the oscillation frequency of which is approximately equal to the center frequency of the corresponding band pass filter; A plurality of selected sine wave signal generation means, and a plurality of amplitude modulation means for amplitude modulating the sine wave signal from a corresponding one of the plurality of sine wave signal generation means based on data from the plurality of level data generation means. , and a synthesizing means for synthesizing the outputs of the plurality of amplitude modulating means. 2. The speech analysis and synthesis device according to claim 1, further comprising a pronunciation means for producing a sound in accordance with the output of the synthesis means. 3. The speech analysis and synthesis apparatus according to claim 1 or 2, further comprising noise level data generation means for detecting the noise level of the voice input in a silent state and generating the level data. 4. Speech analysis and synthesis according to claim 3, wherein the noise level data generation means includes a high-pass filter that receives the voice input, and a noise level data generation means that generates data according to the output level of the high-pass filter. Device. 5. Noise component generation means; Amplitude modulation means for amplitude modulating the noise component from the noise component generation means based on noise level data from the noise level data generation means and providing it to the synthesis means. 4. The speech analysis and synthesis device according to item 4. 6. Claim 3, further comprising voiced/unvoiced determining means for determining whether the voice input is voiced or unvoiced, and controlling the noise component from the noise component generating means in accordance with the discrimination output of the voiced/unvoiced determining means. The speech analysis and synthesis device according to any one of Items 1 to 5. 7. The speech analysis and synthesis device according to claim 5, which includes a voiced/unvoiced discriminator for discriminating whether the voice input is voiced or unvoiced, and controls the amplitude modulation means in accordance with the output of the voiced/unvoiced discriminator. .