GB2103453A

GB2103453A - Speech signal transmission system

Info

Publication number: GB2103453A
Application number: GB08222091A
Authority: GB
Inventors: Peter Bylanski
Original assignee: General Electric Co PLC
Current assignee: General Electric Co PLC
Priority date: 1981-07-31
Filing date: 1982-07-30
Publication date: 1983-02-16
Also published as: GB2103453B

Abstract

A speech signal transmission system in which there are provided a plurality of voice-model speech encoders 3 each arranged in respect of analogue speech signals on a respective input path to provide digitally coded output signals representing during each of a succession of time intervals a model of a vocal tract substantially to reproduce said analogue speech signals, and digital multiplexer means 4 to combine said digitally coded output signals to provide a stream of digits at a rate equivalent to a single sampled-waveform coded speech signal. Where traffic conditions permit said voice-model speech encoders may be arranged to provide said digitally coded output signals two or more times during each basic time interval. <IMAGE>

Description

SPECIFICATION Speech signal transmission systems The present invention relates to speech signal transmission systems.

Speech signals, for example in a telephone system, are commonly encoded for transmission by sampling their instantaneous amplitude at a repetition rate of 8 Kbits/sec and compression-law coding each amplitude sample, for example in accordance with the CCITT A-law, to derive a respective, the code groups being transmitted in succession to give eight-bit pulse code group, an encoded signal bit rate of 64 Kbits/sec. On major links of a transmission path these encoded signals may be interleaved, or multiplexed, with the addition of signalling bits, to provide in one example a single stream of binary digits at a rate of some 140Mbits/sec representing 1920 speech signals. This stream of digits is then transmitted by modulation of 6 carrier signal having a frequency of the order of, say, 6 GHz.

In systems currently being developed, however, speech signals are voice-model coded, for example by means of linear predictive coding analysers, to provide in respect of each speech signal a digitally coded signal at a bit rate of, say, approximately 4 Kbits/sec. Such voice-model coding provides, in respect of each of a succession of 20 m secs "frame" intervals a stream of some 80 bits, denoting, in respect of the nature of the speech signal during that interval, whether it was voiced or unvoiced, its pitch, and a set of electric filter characteristics as accurately as possible mimicking the acoustic characteristics of the vocal tract which produced the speech signal.

According to one aspect of the present invention in a speech signal transmission system there are provided a plurality of voice-model speech encoders each arranged in respect of analogue speech signals on a respective input path to provide digitally coded output signals representing during each of a succession of time intervals a model of a vocal tract substantiallyto reproduce said analogue speech signals, and digital multiplexer means to combine said digitally coded output signals to provide a stream of digits at a rate equivalent to a single sampled-waveform coded speech signal.

Each voice-model speech encoder may comprise an analogue-to-digital convertor and a linear predictvie coder. Alternatively analogue speech signals on said respective input paths may be digitally coded by means a common analogue-to-digital convertor and the digitally coded signals applied to respective linear predictive coders. Each encoder may provide output signals at a rate of 4 or 8 kilobits/sec and sixteen or eight, respectively, of these output signals may be multiplexed to provide a 64 kilobitslsec stream of digits, equivalentto one sampled-waveform speech signal coded in accordance with the CCITT A-law.

According to another aspect of the present invention in a speech signal transmission systen compris ing a plurality of voice-model speech encoders each capable of providing, in respect of an analogue speech signal, digitally coded output signals representing during each of a succession of basic time intervals a model of a vocal tract substantially to reproduce said analogue speech signal, there are provided means arranged selectively to connect one of said encoders to encode one analogue speech signal or more than one of said encoders to encode said one analogue speech signal during overlapping time intervals, to provide up-dating of said coded output signals selectively once per basic time interval or more than once per basic time interval.

According to another aspect of the present invention in a speech signal transmission system there are provided a first plurality of voice-model speech encoders connected to encode signals on respective input paths of a first plurality of input paths during each of a first succession of time intervals, a first digital multiplexer to combine the output signals from said first plurality of encoders, a second plurality of voice-model speech encoders connected to encode signals on respective input paths of a second plurality of input paths during each of a second succession of time intervals, overlapping said first time intervals, a second digital multiplexer to combine the output signals from said second plurality of encoders, and means selectively to interconnect respective first input paths and respective second input paths when speech signal traffic conditions permit whereby encoded signal in respect of any one speech signal may be updated by respective encoders in each of said first and second pluralities in turn.

A speech signal transmission system in accordance with the present invention will now be described by way of example with reference to the accompanying drawing, of which Figures 1 and 2 shows part of a system schematically.

Referring first to Figure 1 speech signals on a number of input channels 1 are applied by way of a conventional analogue-to-digital converter arrangement 2 to individual voice-model coders 3.

The converter arrangements 3 may either comprise individual converters (not shown) for each channel 1 or may be a singletime-shared convertor (not shown). In either case it is arranged to sample the instantaneous amplitude of each speech signal, typically at a rate of 8 KHz, and to digitally encode the amplitude samples in twelve-bit binary form for application to the respective coders 3.

The voice-model coders 3, which may be linear predictive coders, provide digitally encoded output signals at bit rates of, say, 4 Kbits/sec or 8 Kbits/sec indicating parameters of the speech signals on the respective input channels 1. Thus, in each of a succession of 20 msecframe periods the coders 3 provide digit values indicating whether the input speech signal is voiced or unvoiced, the pitch, and the electrical characteristics of a filter required in a subsequent speech synthesizer (not shown) to mimic the accoustic characteristics of the vocal tract which produced the speech signal. At a bit rate of 4 Kbits/sec each coder 3 produces some eighty bits, representing speech information which would normally be conveyed by one hundred and sixty amp litudesamples each compression-law coded to provide a group of eight bits.

The digital output signals from sixteen or eight respectively of the individual coders 3 are combined in a multiplexer 4 to give a 64 Kbits/sec digit stream equivalent to a single amplitude-sampled PCM signal. This 64 Kbit/sec digit stream may be further multiplexed with other streams before transmission.

The one hundred and sixty samples encoded by the arrangement 2 in respect of a speech signal in any one frame period are stored in the respective coder 3 for analysis in the next frame perion. The enelysis or coding process normally occupies most of the 20 msecframe period, although it may be carried out more quickly, depending on the accuracy required or the capabilities of the coder 3.

As shown in Figure 2 the encoders 3 and multiplexers 4 may be arranged in two groups 5 and 6 each having its own set of speech input channels 1 but the channels associated with one group 5 being selectively switchable by way of logic circuits 7 to the inputs of the group 6 in place of the channels normally associated with the group 6.

The frame periods for the coders 3 of the group 6 may be arranged to commence midway through the frame periods for the coders 3 of the group 5, and the respective outputs may simply be interleaved or multiplexed together in output logic circuits 8. When the number of speech channels 1 actually in use is equal to or less than the number of associated with the group 5 these speech channels may be con nected by way of the circuits 7 to both sets of coders, so that in respect of any one active speech channel 1 overlapping sets of speech samples are analysed and the corresponding voice-model coded output signals are made available at intervals of 10 msecs instead of 20 msecs. In the synthesizer (not shown) at the receiving end of the system therefore the voice model may be up-dated twice as often as normal, giving improved quality in the resulting synthesized speech.

It will be appreciated that where the analysis process can be controlled reliably in less than the 20 msecs frame period and traffic conditions permit a more frequency updating scheme other than the above two to one scheme may be possible.

Claims

1. A speech signal transmission system wherein there are provided a plurality of voice-model speech encoders each arranged in respect of analogue speech signals on a respective input path to provide digitally coded output signals representing during each of a succession of time intervals a model of a vocal tract substantially to reproduce said analogue speech signals, and digital multiplexer means to combine said digitally coded output signals to provide a stream of digits at a rate equivalentto a single sampled-waveform coded speech signal.

2. A speech signal transmission system in accor dance with Claim 1 wherein each voice-model coder comprises an analogue-to-digital convertor and a linear predictive coder.

3. A speech signal transmission system in accordance with Claim 1 wherein analogue speech signals on said respective input paths are digitally coded by means of a common analogue-to-digital converter and the digitally coded dignals applied to respective linear predictive coders.

4. A speech signal transmission system in accordance with any preceding claim wherein each encoder provides output signals at a rate of 4 to 8 kilobits/sec and sixteen or eight, respectively, of these output signals are multiplexed to provide a 64 kilobits/sec stream of digits, equivalent to one sampled-waveform speech signal coded in accordance with the CCITT A-law.

5. A speech signal transmission system comprising a plurality of voice-model speech encoders each capable of providing, in respect of an analogue speech signal, digitally coded output signals representing during each of a succession of basic time intervals a model of a vocal tract substantially to reproduce said analogue speech signal, wherein there are provided means arranged selectively to connect one of said encoders to encode one analogue speech signal or more than one of said encoders to encode said one analogue speech signal during overlapping time intervals, to provide updating of said coded output signals selectively once per basic time interval or more than once per basic time interval.

6. A speech signal transmission system wherein there are provided a first plurality of voice-model t speech encoders connected to encode signals on respective input paths of a first plurality of input paths during each of a first succession of time intervals, a first digital multiplexer to combine the output signals from said first plurality of encoders, a second plurality of voice-model speech encoders connected to encode signals on respective input paths of a second plurality of input paths during each of a second succession of time intervals, overlapping said first time intervals, a second digital multiplexer to combine the output signals from said second plurality of encoderso and means selectively to interconnect respective first input paths and respective second input paths when speech signal traffic conditions permit whereby encoded signal in respect of any one speech signal may be updated by respective encoders in each of said first and second pluralites in turn.

7. A speech signal transmission system substantially as herein before described with reference to the accompanying drawing.