JPH02203400A - Voice encoding method - 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] [Field of Industrial Application] The present invention uses a speech encoding method, particularly an adaptive predictive coding method, to encode speech with high efficiency and to encode this encoded speech information at high speed in real time. The present invention relates to a voice encoding method for transmission to a computer.

[Prior Art] There is a wide demand for audio encoding methods for digitally encoding audio in fields such as telephone communications, videophone systems, and voice recognition. Proposed.

FIG. 2 shows an example of a conventional speech encoding device, in which a sample sequence of electrically captured speech is supplied to the speech encoding device from an input terminal 10, and an adaptor U
In order to perform J encoding, subtraction processing with the predicted value is first performed.

For this purpose, the input signal is supplied to the linearizer nj circuit 12, and linear prediction coefficients corresponding to the input signal are set in the predictor 14.

On the other hand, during this predicted value calculation, the human input signal is transmitted to the delay circuit 16.
The predicted value is delayed for a predetermined time by , and is supplied to the subtracter 18 together with the predicted value 7JPj that is the output of the predictor 14, where the predicted value is subtracted from the input signal.

The difference signal, which is the result of the subtraction by the subtracter 18, is frequency-converted by the FFT circuit 20, and the output spectrum information is quantized and encoded by the quantization encoder 22, and this spectrum information is sent from the multiplexer 24 to the output terminal. 26 and output as a digitally encoded audio encoded signal.

Therefore, according to such a voice encoding method, it is possible to digitally encode voice sample value series information in real time and transmit the same.

[Problems to be Solved by the Invention] However, such conventional audio encoding methods have a problem in that the amount of digitally encoded information is limited by the transmission encoding speed during real-time transmission.

For example, when transmitting at the 3Kbps encoding speed used in normal transmission, the number of quantization bits of the frequency-converted differential signal described above decreases, and the quality of the reproduced audio on the receiving side deteriorates significantly. There were drawbacks.

The present invention has been made in view of the above-mentioned conventional problems, and an object of the present invention is to provide an improved speech encoding system that can maintain the quality of received encoded speech at a sufficient level despite the limitations of transmission speed. The purpose is to provide a method.

[Means for Solving the Problems] In order to achieve the above object, the present invention performs quantization encoding with emphasis on the formant frequency and frequency components in its vicinity when encoding a frequency-converted difference signal. The present invention is characterized in that this formant encoded signal is transmitted alone or together with the linear prediction coefficient obtained by the linearizer n1 circuit, thereby making it possible to transmit encoded speech information of sufficiently excellent quality even with a small number of bits. do.

[Operation] Therefore, according to the present invention, even when there is a restriction on transmission speed, signals in the formant frequency band that give the characteristics of vowels in speech are transmitted preferentially, and other low-density information is cut. This makes it possible to transmit high-quality encoded audio at normal transmission speeds without degrading quality.

[Example] Preferred embodiments of the present invention will be described below based on the drawings.

FIG. 1 shows an example of an encoding circuit to which the speech encoding method according to the present invention is applied, and the same members as those in the conventional circuit shown in FIG. do.

As in the past, in the present invention, electrically captured voice sample value sequence information is manually inputted from the input terminal 10, and then framed with a specific length (for example, 20 to 30 m5ec), and then sent to the linear prediction circuit. At step 12, a prediction coefficient is calculated, and a difference value between the sample tenitor and the predetermined value, which is the output of the predictor, is extracted. After converting the frequency of this difference value, only the information of a specific spectrum is quantized, encoded, and transmitted, thereby making it possible to significantly reduce the number of quantization bits and achieving highly efficient speech encoding.

The characteristic feature of the present invention is that the difference signal frequency-converted by the FFT circuit 20 is not entirely quantized and encoded as in the past, but only a specific frequency component is quantized and transmitted. , FFT circuit 20
A frequency thinning circuit 10 is provided between the
0 is set.

Therefore, the spectral information is transmitted to this frequency thinning circuit 100.
Only the frequency band defined by is selectively supplied to the quantization encoder 22, and if the selected frequency components are set only to bands that are important for the quality of reproduced audio, it can be used even at relatively low transmission speeds. It becomes possible to accurately transmit important audio information.

In the present invention, the selected frequency band is a formant frequency that is a resonance frequency on the audio spectrum and a band adjacent thereto.

As is well known, formant frequencies are frequency bands in which dominant components that characterize vowels exist, and are usually called first, second, third formant frequencies, etc. in order from the lowest frequency.

Therefore, in this embodiment, the formant calculation circuit 101 calculates the formant bandwidth for the voice currently being processed from the output of the linear prediction circuit 12, and thereby the frequency thinning circuit 100 calculates the frequency component near the formant frequency of the difference signal. The spectral information to be subjected to quantization and encoding is selected by performing a filtering action that only allows the spectral information to pass through.

In practice, the formant bandwidth for passing spectral information can be calculated by the formant calculation circuit 101 from the linear prediction coefficients as described above, and the spectral position and number of lines that need to be encoded corresponding to this formant bandwidth can be calculated using the formant calculation circuit 101. A table that corresponds on the frequency axis is provided inside the formant calculation circuit 101 or the frequency thinning circuit 100, thereby making it possible to provide only necessary formant band signals to quantization encoding in real time.

Of course, a table similar to the above-mentioned conversion table is also provided on the decoder side, so that the formant calculation circuit on the decoder side can perform decoding without sending position information on the frequency axis of the transmitted spectrum in the same way as the encoder. Accurate decoding can be easily performed by retrieving the bandwidth from the table.

Assuming that the normal audio band is about 0.3 to 3.4KH2,
It is only necessary to consider about three formant bands, first to third, and by deleting the other spectra, it is possible to significantly reduce the number of spectra that need to be encoded.

Therefore, it is possible to increase the number of quantization bits per piece of sampling information in the difference signal frequency-converted in this manner, and as a result, it is possible to significantly improve the audio quality.

Further, according to the encoding method of the present invention, only the differential signal selected in the formant frequency band is sent even at a relatively low transmission rate, for example, about 3 kbps, so transmission with little quality deterioration can be achieved. possible in real time.

In the embodiment, the formant information used in the frequency thinning circuit is updated for each frame and is therefore changed by the predicted value, and for this reason, the preliminary value of the linear prediction circuit 12 is also transferred from the multiplexer 24 to the quantization It is output from the output terminal 26 to the transmission system together with the encoded difference information.

However, in the present invention, if the formant band is fixed within a region covering a specific or all vowels, the output of the prediction circuit 12 does not necessarily need to be sent to the transmission system.

[Effects of the Invention] As explained above, according to the present invention, in the speech encoding method in which only the difference signal is transmitted by the adaptor ΔP1,
By further transmitting the difference information as information only in the formant frequency band, it is possible to perform high-density speech encoding.

Therefore, according to the present invention, even if the transmission speed is limited, voice quality deteriorates, and if a sufficient transmission speed is allowed, encoded transmission with excellent reproducibility is possible. That is understood.

[Brief explanation of the drawing]

FIG. 1 is a circuit explanatory diagram showing a preferred embodiment of an encoding circuit to which the audio encoding method according to the present invention is applied, and FIG. 2 is an explanatory diagram of a conventional audio encoding circuit. 12 ... 20 ... 22 ... 100 ... 101 ... Linear element 11111 circuit FFT circuit Quantization encoder Frequency thinning circuit Formant system calculation circuit

Claims (1)

[Claims] In an audio encoding method that calculates a difference between audio information using each sampling value and its predicted value, quantizes and encodes the difference information, and transmits the same, in which spectrum information in a formant band of the difference signal is selectively encoded and transmitted. A speech encoding method characterized by: