Speech Signal Transmission Systems Utilizing A Non-linear Circuit In The Base Band Channel

Abstract

In a speech signal transmission system there are provided means including a non-linear circuit and a bandpass filter to produce from a speech signal a base band signal representing the spectral fine structures of the speech signal and to transmit the base band signal to the receiving side, a vocoder channel analyzer to convert the speech signal into a second signal representing the spectral envelope of the original speech signal, a second non-linear circuit on the receiving side to convert the base band signal into an exciting signal and a vocoder synthesizer which acts to synthesize the original speech signal from the exciting signal and the second signal.

Description

BACKGROUND OF THE INVENTION

This invention relates to a speech signal transmission system and more particularly to a speech signal transmission system employing a novel speech band compression system.

Band compression of speech is important for communication systems utilizing expensive transmission circuits, such as communication systems utilizing satellites or submarine cables. A channel vocoder is a typical speech band compression system. According to this system, the frequency spectrum of the speech signal is analyzed into signals or informations representing the spectral envelope and fine structures thereof on the transmission side, the spectral envelope is detected by more than 10 bandpass filters, rectifiers and low pass filters while spectral fine structures are detected by determining whether the sound is a voiced sound or a unvoiced sound and by extracting the pitch frequency in the case of the voiced sound. The informations regarding spectral envelope and spectral fine structures are sent from the transmission side as a plurality of band compressed signals. On the receiving side a frequency spectrum approximating that of the original speech is reproduced from these signals. While this system provides a band compression ratio of more than ten to one it is difficult to detect the spectral fine structures of the speech on the transmission side thus lacking articulation and naturalness in the reproduced speech.

To eliminate these difficulties, a voice excited vocoder (VEV) has been developed wherein determination of the spectral fine structures is not performed on the transmission side. According to this system a portion of the speech band near the lower end of the speech spectrum (hereinafter termed as the base band) is transmitted directly to the receiving side thus eliminating the necessity of the detection of the voiced sound and unvoiced sound as well as the extraction of the pitch frequency.

FIG. 1 shows diagrammatically the construction of this system according to which components in the base frequency band f.sub.1 to f.sub.3 (f.sub.1 <f.sub.3 <f.sub.2) of the speech signal ranging from f.sub.1 to f.sub.2, where f.sub.1 <f.sub.2, and impressed upon an input terminal 1 are separated by a bandpass filter 2 of the frequency band f.sub.1 to f.sub.3 and the separated components are transmitted to the receiving side as the base band signal through one of the transmission lines 4. The remaining components of the input speech in the frequency band f.sub.3 to f.sub.2 are converted into a plurality of signals with their frequency bands compressed by a vocoder channel analyzer 3, and the converted signals are transmitted to the receiving side over other transmission lines.

On the receiving side, the received base band signal is sent to an output terminal 8 through an adder 7. The received base band signal is also supplied to a non-linear circuit 5 to regenerate, by the action of the non-linear circuit, components in the frequency band f.sub.3 to f.sub.2 which have been removed on the transmission side, thus providing an exciting signal in the frequency band f.sub.1 to f.sub.2 containing fine structures of the original speech spectrum. Signals from the vocoder channel analyzer 3 are supplied to a vocoder synthesizer 6 and combined therein with the exciting signal to reproduce components of frequency band f.sub.3 and f.sub.2 of the original speech. The reproduced components are sent to the output terminal 8 via adder 7 thus reproducing all components of f.sub.1 to f.sub.2 of the original speech at the output terminal 8.

As above described since in the VEV system a portion of the original speech spectrum is transmitted without being processed in any way, the qualities of the reproduced speech, such as articulation and naturalness, are excellent but this system requires wide transmission band thus decreasing the band compression ratio. The band width required to transmit the base band signal is determined in the following manner. More particularly, since the base band signal serves to transmit informations of the pitch frequency;

1. As long as the pitch frequency is included in the base band, the base band is not required to contain higher harmonic components of the pitch frequency. 2. Where the pitch frequency is not included in the base band it is necessary that at least two adjacent higher harmonic components of the pitch frequency should be included in the base band.

The pitch frequency of ordinary speech generally ranges from about 50 to 450 Hz so that the base band always satisfying either one of the two conditions mentioned just above is determined in the following manner. Denoting the pitch frequency of the speech by f.sub.0, the lower limit of its variation by f.sub.01, the upper limit by f.sub.02, base band by f.sub.L to f.sub.U (where f.sub.L <f.sub.U) and its band width by f.sub.B (= f.sub.U -f.sub.L) then

from condition 1 : f.sub.L .ltoreq.f.sub.0 .ltoreq.f.sub.U

from condition 2 : 2f.sub.0 .ltoreq.f.sub.U -f.sub.L .ltoreq.f.sub.B

With reference to FIG. 3, the shaded area shows the range of f.sub.B which satisfies at least one of thess conditions. The solid line in FIG. 3 shows the necessary minimum value of f.sub.B for a given f.sub.L, or the lower limit of the base band, when f.sub.02 is greater than 3f.sub.01 as in the case of conversational speech of an indefinite number of talkers covering a wide range of variations in the pitch frequency. In this case, when f.sub.L is selected to be equal to 1/3f.sub.02 and f.sub.U is selected to be equal to f.sub.02 the base band width will be minimum and thus the minimum value 2/3f.sub.02 is obtained. Where a particular talker is specified, the range of the pitch frequency would be narrowed to satisfy the condition of f.sub.02 <3f.sub.01, so that a small shaded triangular range 9 shown in FIG. 3 will be added with the result that the minimum value of f.sub.B is realized when f.sub.L is selected to be qual to f.sub.01, and f.sub.U is selected to be equal to f.sub.02 and thus the minimum value f.sub.02 minus f.sub.01 is obtained. As above described since the pitch fr950000000000000000000000000000000000000000000000000000000000000000