JPS59172690A - Vocoder - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] [Technical field to which the invention pertains] The present invention relates to a vocoder (
ν0CODER).

[Description of Prior Art] □ Vocoders have been researched with the aim of compressing the voice band to a fraction of that of a normal telephone line to several tenths of that of a normal telephone line, and making effective use of the line, and some have even been commercialized. There is.

With the recent development of digital signal processing technology, vocoders have also been digitized, and recent ones have been put into practical use for transmitting audio at 2400 bits/second to 4800 bits/second. The mainstream method is the partial autocorrelation coefficient (PAR
This is called a linear predictive vocoder that uses COR coefficients as spectrum analysis information.

In a linear predictive vocoder, a transmitter calculates and transmits a PARCOR coefficient, a residual power, a pinch period, and a voiced/unvoiced determination result. In the receiving section, a synthesis filter is configured based on the sent PARCOR coefficients, and when voiced, a pulse with an amplitude corresponding to the residual power is input according to the pitch period, and when unvoiced, a white pulse with an amplitude corresponding to the residual power is input. It has a structure that inputs noise and generates a combined voice.

FIG. 1 shows the block configuration of such a conventional vocoder. In the same figure, the PARCOR coefficient is found in the spectrum analysis circuit 1, and the pinch detection circuit 2
A pinch is detected at . The output of the voiced/unvoiced determination circuit 3 is used on the receiving side to select whether to use the output of the white noise generator 7 or the output of the pulse generator 8 as an excitation source for the synthesis filter 9. The magnitudes of the outputs of the self-noise generator 7 and the pulse generator 8 are controlled by the residual power determined by the spectrum analysis circuit 1. The period of the output pulse of the pulse generator 8 is determined by the output of the pinch detection circuit.

With the above-described configuration, the vocoder clearly distinguishes between voiced and unvoiced sounds, and therefore, a decision error made by the voiced/unvoiced decision circuit 3 causes quality deterioration of the synthesized speech.

However, in human speech, intermediate voices that are difficult to determine as voiced or unvoiced also exist at the boundary between vowels and consonants, and when this boundary is determined to be voiced, the pinch detection circuit 2 accurately detects the pitch. is difficult, and if the voice is determined to be voiceless, the synthesized sound will sound hoarse. In the case of this intermediate voice, it is considered that white noise and pulses are simultaneously added and input on the receiving side, but as mentioned above, it is difficult for the pinch detection circuit 2 to accurately detect the pitch. Since it's just a voice, it doesn't really help improve the sound quality.

[Purpose of the invention]

The present invention solves the above-mentioned drawbacks, and aims to provide a vocoder that improves the quality of synthesized sounds on the receiving side of sounds that are difficult to determine as voiced or unvoiced.

[Key points of the invention]

When transitioning from voiceless to voiced, the vocal cords often produce isolated pulses or pulses with irregular periods, with the vibration period still being constant. This also causes pitch errors to be detected.

Therefore, in order to improve quality, it is desirable to send the waveform itself rather than the pinch cycle. Adaptive predictive coding is used as a waveform transmission method.
tive coding), the remainder after linearly predicting the input signal, that is, the residual signal, is transmitted,
Ensures high quality.

Based on this, it is expected that the quality can be improved by sending the residual waveform.

In the case of the above-mentioned adaptive predictive coding, the transmission rate is 16K.
Since the speed is B~24 KB/sec, which is several times that of a vocoder, it cannot be applied to a vocoder as is.

Therefore, it is possible to obtain a certain level of quality by sending an outline of the residual waveform. In addition, in order to avoid increasing the transmission rate by sending the residual waveform, the part that sends bits for pitch information during silent periods and part of the spectrum information is used to send the residual waveform instead of these information. Send the difference waveform. Originally, it seems that the temporal waveform rather than the spectrum information is more important in the transient part, so if the spectrum information is halved, for example, an outline of the residual waveform can be sufficiently transmitted.

Based on the above considerations, in the present invention, when it is determined that there is no voice, the PARCOR coefficient is sent only about half the order of the voiced one, and the bit allocation for the pitch and the bit allocation for the pitch (pinch) are sent. is not needed when there is no voice) and is used for residual waveform transmission. For this purpose, a residual waveform synthesis circuit is provided to input the PARCOR coefficients determined by the spectrum analyzer to obtain a residual waveform, and a transmission signal switching circuit is used to enter the determined residual waveform into the bit allocation described above. This switching circuit functions to send out the PARCOR coefficient and pitch when there is voice, and to send out the residual waveform when there is no voice.

In addition, in the receiving section, the received signal switching circuit switches based on the received voiced/unvoiced determination result, and when voiced, the PARCOR coefficient and pinch information are used as a synthesis filter and a pulse is generated, respectively, just like in a conventional vocoder. On the other hand, when there is no voice, the residual waveform is added to the output of the zero noise generator.

In order to realize the above-mentioned effects, the present invention is characterized in that the transmitter includes a residual waveform generation circuit, a voiced/unvoiced determination circuit, a pitch detection circuit, first and second transmission signal switching circuits, and a spectrum. an analysis circuit, the output of the voiced/unvoiced determination circuit is connected as a control input of the first and second transmission signal switching circuits, and the output of the pinch detection circuit, the output of the spectrum analysis circuit, and the residual waveform are connected. The output of the combining circuit is connected as the input of the first and second transmission signal switching circuits,
The reception section includes first and second reception signal switching circuits, a noise generator, a pulse generator, a synthesis filter, an input switching circuit, and an adder, and the transmission signal from the transmission section is connected to the reception signal switching circuit. The output of the switching circuit is connected, and the output of the voiced/unvoiced determination circuit is connected to the control input of the first and second received signal switching circuits and the control input of the input switching circuit, One output of the signal switching circuit becomes the input of the adder, the other output of the first received signal switching circuit becomes the characteristic input of the synthesis filter, and the second
The other output of the received signal switching circuit becomes the input of the pulse generator, the output of the noise generator is connected to the input of the adder, and the output of the adder is connected to the input of the input input switching circuit. The output of the pulse generator is connected as an input of the input switching circuit, and the output of the input switching circuit is connected as an input of the synthesis filter.

[Explanation based on examples]

Embodiments of the present invention will be described below based on the drawings.

FIG. 2 is a block diagram showing a device according to an embodiment of the present invention, in which a spectrum analysis circuit 1, a pinch detection circuit 2, a voiced/unvoiced determination circuit 3, a noise generator 7, a pulse generator 8, and a synthesis filter are shown. 9 is the same as that in FIG. The residual waveform synthesis circuit 4 forms an inverse filter based on the information from the spectrum analysis circuit 1, and generates a residual waveform from the input signal. This residual waveform is not transmitted when voiced, but when unvoiced, a part of the icOR coefficient (for example, the 3rd to 10th order part) and pinch information are transmitted by the transmission signal switching circuit 5, which is switched and controlled by the voiced/unvoiced determination circuit 3. transmitted instead.

On the receiving side, the received signal switching circuit 6 is driven by the voiced/unvoiced determination result, and the residual waveform is distributed to the adder 10, the PARCOR coefficient to the synthesis filter 9, and the pinch information to the pulse generator 8.

FIG. 3 shows the transmission format used in the embodiment of the present invention, in which A represents the one-frame format when voiced and B represents the one-frame format when voiceless. 31 is a bit for frame synchronization, 32 is a PARCOR coefficient from the 1st to the 10th order,
33 is the same < P/1. Rc (first to third JR coefficients, 34 is a residual waveform, 35 is a pinch, 36 is a residual power, and 37 is a voiced/unvoiced determination bit.

Next, the operation of this embodiment device will be explained.

When the input voice is determined to be unvoiced by the voiced/unvoiced determination circuit 3, the P obtained by the spectrum analysis circuit 1
The ARCOR coefficient only sends about half the orders (for example, 1st to 3rd) when voiced, and the bit allocation for the remaining orders and the bit allocation for pinch information detected by the pinch detection circuit 2 (pinch is (not required when there is no voice) is used for transmitting the residual waveform generated by the residual waveform synthesis circuit 4. That is, in order to put the residual waveform obtained by the residual waveform synthesis circuit into the above-mentioned hit assignment, a transmission signal switching circuit 5 is used, and the switching of this switching circuit 5 is performed according to the output of the voiced/unvoiced judgment circuit 3. When there is voice, the PAIICOR coefficient and pinch information are sent to the receiving section via the switching circuit 5, and when there is no voice, the bit allocation residual waveform mentioned above is sent out.

In addition, in the receiving section, the received signal switching circuit 6 is switched based on the received voiced/unvoiced determination result, and if voiced, the received signal switching circuit 6 is switched in the same way as a conventional vocoder: On the other hand, when there is no voice, the residual waveform is added to the output of the internal conical sound generator 7 by an adder 10.

In addition, in the above embodiment apparatus, for convenience of explanation, the PARCOR coefficient was used as an example of the analysis parameter, but the present invention is equally applicable even when other analysis parameters such as LSP (line spectrum pair) are used. It is possible.

In addition, the formant example in Figure 3 shows a fixed length frame in which the frame length does not change between voiced and unvoiced times, but this is a variable length frame in which the frame length is lengthened during unvoiced times and the residual waveform is sent accurately. It goes without saying that the same effect can be obtained by using .

〔Effect of the invention〕

As explained above, in the present invention, the quality deterioration caused by the conventional vocoder using only white noise as an excitation source during unvoiced periods can be overcome by taking advantage of the fact that there is no need to send detailed spectral information during transition periods. This can be improved by using the residual waveform in addition to white noise as the excitation source.
This allows the clarity to be dramatically improved.

This can be understood from the fact that the distinction between the initials of sounds, other than vowels, is highly dependent on the transitional part.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a conventional vocoder. FIG. 2 is a professional configuration diagram showing an apparatus according to an embodiment of the present invention. FIG. 3 is a diagram showing the format of a transmission frame used in the embodiment device of FIG. 2. ■... Spectrum analysis circuit, 2... Pinch detection circuit, 3... Voiced/unvoiced judgment circuit, 4... Residual waveform synthesis circuit, 5... Transmission signal switching circuit, 6... Reception signal Switching circuit, 7... Self-noise ordering device, 8... Pulse generator, 9... Synthesis filter, 10... Adder. Patent applicant: NEC Corporation Representative Patent attorney: Naotaka Ide

Claims (1)

[Claims] (1) The transmitter includes an analysis circuit that analyzes the spectrum of input audio, a detection circuit that detects the pinch period of input audio, and determines whether the input audio is voiced or unvoiced based on information from the analysis circuit and detection circuit. The receiving section includes: a noise generator whose output level changes based on the information from the analysis circuit mentioned above; and a noise generator whose output level changes based on the information from the analysis circuit mentioned above. A pulse generator whose output pulse period is determined based on the pitch information from the detection circuit, and a judgment signal from the judgment circuit described in 1. Switching is controlled to control the output direction of the pulse generator when there is voice, and to generate the noise when there is no voice. In the vocoder, the vocoder is equipped with a switching circuit that selects and outputs a signal output, and a synthesis filter that generates a synthesized sound using the switching circuit output as an excitation source with characteristics determined based on information from the analysis circuit. The unit includes a synthesis circuit that generates a residual waveform signal from the input audio based on the information from the analysis circuit, and the reception unit converts the residual waveform signal sent from the synthesis circuit into the noise signal. It is equipped with an addition circuit that adds to the output of the generator, and when the above judgment circuit judges the input sound to be unvoiced,
The residual waveform signal information is transmitted to the receiving section in place of the pitch information from the detection circuit, and the receiving section is configured to add this residual waveform signal to the output of the noise generator using the adding circuit. A vocoder characterized by: