DE2253222A1 - TRANSMISSION SYSTEM IN WHICH THE LANGUAGE IS TRANSMITTED IN THE FORM OF SAMPLE VALUES - Google Patents

Abstract

1347794 Speech transmission STANDARD TELEPHONES & CABLES Ltd 4 Nov 1971 51276/71 Heading H4R In a speech transmission system formant frequency peaks are reduced in the signal spectrum to be transmitted, e.g. using a self adaptive filter, and are reinserted after transmission, using at the receiver only the transmitted residual signal. As described with respect to Fig. 1 a self adaptive transversal filter 3 is connected in feed forward configuration with the filter target signal equal to the input signal so that the output signal from the subtractor is the input speech signal with formani frequency resonances reduced. A similar filter 4 connected in feedback configuration restores the original speech at the receiver. The trans mission link can include a p.c.m. coder or delta modulator in which case the receiver circuit tends to emphasize quantizing noise. A modified circuit, Fig. 4, in which a decoder 11 provides the input to the coder from the coded output to feed back the quantizing noise, reduces the quantizing noise in the output.

Description

Patent attorney
Dipl.-Phys. Leo Thul -

Stuttgart

L. S. Moye-5

INTERiNlATIONAL STANDARD ELECTRIC CORPORATION, NEW YORK

Transmission system in which the speech is transmitted in the form of samples .

In order to be able to use a transmission system in which the speech is in a form of sampled values is transmitted, it has already been proposed to save transmission bandwidth (DT-OS 22 29 149), that on the transmission side the language is initially a PCM coding is subject to the fact that weighting parameters determined therefrom from the respective preceding PCM signals with the aid of a computer the redundant information is determined that the redundant information is subtracted from the PCM signal and that residual signal produced in this way is transmitted with delta modulation so that the weighting parameters are encoded as a further signal are transmitted in time division multiplex with the residual signal and that on the receiving side of these two signals the original PCM value restored and from this in a known manner the language is regained.

The invention is based on the object of creating a transmission system in which the speech is in the form of sample values is transmitted and in which even less information has to be transmitted than in the case of the proposed

10/17/1972
Ti / Mr

300113/1043 "

L.S. Moye-5

Order. This is achieved according to the invention in that from the sample values to be transmitted in the transmitter in a first Arrangement, the components of the resonance frequencies are extracted and only the remaining signal is transmitted and in the receiver in a second arrangement again the same resonance frequencies are added and that the second Arrangement receives the transmitted residual signal as an input signal. Further features of the invention are set out in the subclaims refer to.

üie invention will now be based on the in the enclosed Drawings illustrated embodiments explained in more detail. Show it:

Fig.l the basic principle of a voice transmission system with unfolding and folding of the speech signals

FIG. 2 shows a modified form of the system according to FIG

3 shows the basic circuit of a self-adapting filter

4 shows an embodiment of a speech transmission system with self-adapting filters for unfolding and convolution of the speech signals and

Figures 5 and 6 are examples of speech signals transmitted over a transmission system according to the invention are sent.

When producing speech, the larynx 1 (Fig. 1), which can be viewed as a pulse generator, excites the pharynx and oral cavity resonators which can be represented as recursive filter 2. If this combination of im-

0 9 819/1043

- 3 - ■ ■

L.S. Moye-b

pulse generator and recursive filter were a complete model, then a counterpart to this filter, i.e. a identical non-recursive filter 3 extract the resonances, or unfold the speech signals, and only those Excitation impulses would remain. That means in other words, that the content of a speech sample is predicted within certain limits from a comparison with previous samples can be. So if it is possible to extract the values from samples of a speech signal that can be predicted, and then only the remaining signal is encoded for transmission, then a achieved a considerable increase in the signal to noise ratio for the actual transmission part of the system will. A signal with a smaller amplitude must then be coded and the quantization errors are corresponding reduced. The part that was pulled out at the sending point of the system is then passed through in the receiving point an arrangement has been added again which is the counterpart of an unfolding device, namely a folding device. Such an arrangement is shown schematically in FIG. 1, in which the convolution device has a recursive filter 4, which corresponds to the recursive filter of the model of the voice channel.

There is an obvious objection to such a system in that the quantization errors in the coded part of the system can be enlarged by the recursive filter and that the final amount of quantization noise is nearly is the same size as in a conventional PCM transmission system.

To avoid this disadvantage it is necessary to use the Feed back quantization error and feed it back to the input

309319/1043

L.S. Moye-5

signal of the recursive filter to be added. Such a modified one The system is shown in Fig. 2, in which the output signal of the encoder 6 is applied to a local decoder 5, so that one can get the quantization error. The output signal of the encoder then becomes the input signal at the filter input added.

The resonances of the pharynx and oral cavity vary from one speech sound to the next. To suppress them, needed a transversal filter that adapts when the resonances change. In British patent specification 1 184 653 (= DT-OS 15 37 626) is already a self-adapting one Filter described which adapts so that its output signal corresponds to a desired signal.

Basically, a self-adapting filter is a filter whose output signal is the weighted sum of signals, derived from the input signal and that Vweighting adjusts so that the output signal is as close as possible to the desired signal comes. A simplified illustration of a self-adapting filter is shown in FIG. 3. The input signal χ is applied to a tapped delay line that has delay sections of St. The output signals of the delay line are applied to multipliers, the weightings of which are a1 to an accordingly the adaptation rules that are specified by the desired signal Z are determined. The output signal y, i.e. the sum of the multiplied output signals should then equal the desired signal ζ.

In the present application, the desired signal and the input signal daselbev Da are only delayed input signals for use in the output signal, it adapts it by taking the best approximation for the current samples from previous sampled values of the input signal

309819/1043 "'"

L.S. Moye-5

directs. Provided that the self-adapting Filter 3 in the transmitting station can adapt quickly enough, it pulls out the resonances that occur in the speech -, - If This filter extracts the resonances from the language, only a small amount of information needs to be included in the resultant Signal can be coded. This gives a reduction in the quantization noise compared to the coding of a normal speech signal. When the resonances are added back in the receiver, it becomes the original input signal restored, but containing the (reduced) quantization errors. These errors are also over routed through the convolver and this would increase the size of the errors considerably. To prevent this the non-recursive filter in the transmitting station is fed with the decoded output signal of the encoder and not with 'the original input signal. It will then only produce that -. pulled out that in the receiving point through the recursive filter can be reinserted.

A system constructed in this way is shown in FIG. The input signal is applied to a dynamically companding 2-bit PCM coder 10, the output signal of which is both to a local decoder 11 as well as to the decoder 12 of the receiver. The output of the decoder is combined with the output signal of the self-adjusting filter (SAF) 13 is applied to the input of this self-adjusting filter 13. This combined signal also serves as a desired signal for the aslb-adapting filter. The output signal of the self-adapting filter 13 is also subtracted from the input signal of the encoder 10. In the receiver the output becomes signal of the decoder 12 as an input signal and desired

L.S. Moye-5

Signal applied to the self-adapting filter 14, which is constructed as a recursive filter. The output of this The filter is added to the output of the decoder.

In Figures 5 and 6, the operation of the system is according to Fig. 4 when transferring the beginning and the end of the word "yes" is explained by the signals occurring at various points in the arrangement according to FIG. Though differences exist between the original input signal and the output signal, they are insignificant compared to the differences that exist between the input signal of the encoder and the output signal of the decoder. Reducing the

Note the bandwidth required for coding the unfolded signal, enjib it
is more substantial bandwidth that would be necessary to directly encode the input speech signal. In a system with a line transmission speed of 9.6Kb / S, the quality of the output speech signal is comparable to the signal obtained from an existing 19.2Kb / S companding delta modulation system.

As an alternative to a 2 bit PCM transmission system between the transmitter and the receiver can also have other digital connections used for low speed, such as delta modulation.

309819/1043

Claims (4)

- 7 -
LS Moye-5 Claims Cly transmission system in which the speech is transmitted in the form of samples, characterized in that the components of the resonance frequencies are extracted from the samples to be transmitted in the transmitter in a first arrangement (TFl) and only the remaining signal is transmitted and in the receiver in the same resonance frequencies are again added to a second arrangement "(TF2) and that the second arrangement only receives the transmitted residual signal as an input signal. 2. Transmission system according to claim 1, characterized in that dfe arrangements for pulling out and inserting the resonance frequencies are transversal filters and that the filter of the transmitting station is designed as a non-resursive and the filter of the receiving station is designed as a recursive filter. 3. Transmission system according to claim 2, characterized in that the transversal filters are identical self-adapting filters, the input signals of which are aubh the desired signals. 4. Transmission system according to claim 1, characterized in that a decoder (11) is provided for a PCM transmission link in the transmission point, to which the output signal of the encoder (10) is applied, that the output signal of the decoder together with the output signal of the filter (13) is applied as an input signal and as a desired signal to the filter that the output signal of the filter is subtracted from the input signal of the encoder and that in the receiving stile the output signal of the filter (1 * 4) is added to the output signal of the decoder (12) and represents both the output signal of the receiver and the input signal and the desired signal for the filter. 309819/10 4 3.