US3431355A

US3431355A - Device for excitation controlled smoothing of the spectrum-channel signals of a vocoder

Info

Publication number: US3431355A
Application number: US445563A
Authority: US
Inventors: Ernst H Rothauser; Erwin F Paulus
Original assignee: International Business Machines Corp
Current assignee: International Business Machines Corp
Priority date: 1965-03-25
Filing date: 1965-04-05
Publication date: 1969-03-04
Anticipated expiration: 1986-03-04
Also published as: AT251656B; BE677136A; SE323721B; NL6602974A; GB1073203A; ES324621A1; CH443413A; DE1276740B; FR1471628A; NL146672B

Abstract

1,073,203. Vocoders. INTERNATIONAL BUSINESS MACHINES CORPORATION. Match 15,1966 [March 25, 1965], No. 11188/66. Heading H4R. Channel. vocoder in which each spectrum analysis channel includes a band-pass filter, a rectifier, and a low-pass filter, or integrator circuit; has a voiced-unvoiced detector the output from which is used to lower the cut-off frequency of the low-pass filter, or increase the integration time of the integrator circuit, when unvoiced sounds are detected; in order to reduce noise in the channels. The voiced-unvoiced detector may cause switching between two alternative filters provided for each channel, Fig. 2 (not shown), or it may modify the filtercircuit, e.g. by operation of switch Si. in the filter of Fig. 3. A typical integrator circuit which might be used in each spectrum analysis channel is shown in Fig. 5 in which the spectrum channel signal is applied to band causes capacitor C to be discharged by an amount proportional to the current applied through R 1' Periodically a signal is applied to line a to cause TR 2 to conduct to charge C up to the supply potential. The charging current, which will depend on the total discharge of C since the last charging pulse, flowing through R3 will produce an output pulse on Ki<SP>1</SP> the amplitude of which is proportional to the integral of the spectrum channel signal over a time equal to the period of the pulse signal applied to line a. The voiced-unvoiced discriminator output is employed to increase the period between the pulses applied to line b during unvoiced sounds.

Description

March 4. 1969 E. H. ROTHAUSER ET AL 3,431,355

SMOOTHING OF T A VOGODER Sheet 3 of 2 DEVICE FOR EXCITATION CONTROLLED SPECTRUM-CHANNEL SIGNALS OF Filed April 5, 1965 FlGQi FIG. 2

0 II n In K K K K u u m u u L 2 pm B 2 n 1 S s w :llii- Illii I 2 '2 n n A A A A A A A... Ill 2 2 n \n P P P P P w L L L L 2 n m G G 1 2 n P P P B B B O 2 n M K K K N m Filed April 5, 1965 E. H. ROTHAUSER ET AL DEVICE FOR EXCITATION CONTROLLED SPECTRUM-CHANNEL SIGNALS OF 3,431,355 SMOOTHING OF THE A VOCODER Sheet 2 ,LP Li Li' vm n'm Ai Ail FIG.4

U K b 1 SP1 G1 A i K1 I 'i Al K2 b I Kl 8P2 G2 A2 2 i 1 52 lo I 1 1 2. 3 K b 5P 6,, In n J An 1 KO United States Patent A 2,756/ 65 U.S. Cl. 179-1 Int. Cl. H04m 1/00 5 Claims ABSTRACT OF THE DISCLOSURE Disclosed is a method and apparatus for achieving a reduction of the noise component of the channel voltages in the operation of a channel vocoder during unvoiced speech segments without impairing the quality of voiced segments. The output of each channel is limited to a frequency which is lower for unvoiced segments than for voiced segments. This can be achieved by inserting a variable filter arrangement in each channel for smoothing the output of the same according to a lower limiting frequency for unvoiced rather than for voiced information segments. The filter arrangement may consist of a pair of low pass filters for each channel, or a variable cut-off low pass filter for each channel, or a triggered integrator for each channel operable in a higher triggering rate or lower triggering rate mode, in each case the filter arrangement being switched between its higher cut-off and lower cutoff mode by the voiced-unvoiced discriminator of the vocoder.

This invention relates to methods and arrangements for reducing the noise components contained in the channel voltages when analyzing unvoiced speech segments.

As is well known, one individual channel of the analyzer of a channel vocoder consists of a band-pass filter BP to BP (FIG. 1), a rectifier G to G and a smoothing lowpass filter LP to LP respectively. When analyzing voiced or unvoiced longerlasting speech segments, the output voltages of the smoothing low-pass filters LP to LP should be constant D.C. voltages. This ideal case is easily obtainable for voiced longer-lasting speech segments. For unvoiced longer-lasting speech segments, on the other hand, the output voltages also contain noise components the energy spectrum of which corresponds to the band width of the smoothing low-pass filter. A noise component gains increasing importance with respect to the constant DC. voltage with any increase of the band width of the smoothing low-pass filter LP, with respect to the band width of the band-pass filter BP In view of the unfavorable etfects of the noise components of the spectrum signals on the quality of the synthetic speech, however, this invention has for its object to reduce these noise components in the spectrum signals when analyzing unvoiced speech segments.

Concerning a method of improving the quality of speech in connection with the analysis of unvoiced speech seg' ments according to the channel vocoder principle, the invention consists in the provision that the noise component of the D.C. signal carried by each spectrum channel after the analysis is reduced by selecting an optimally low value for the upper limiting frequency of the smoothing low-pass filter provided for each spectrum channel in the presence of unvoiced speech segments.

An arrangement for carrying through the above indicated method shows very favorable results due to the provision that each spectrum channel contains two lowpass filters with different upper limiting frequencies both of which are connected, on the one hand, to the associated D.C. circuit and, on the other hand, to a respectively difierent terminal of a controllable switch the output of which carries the respective channel voltage and the control of which is elfected by the output signal of a discriminator provided in a known manner in the excitation channel for the voiced/unvoiced discrimination of the speech signal, in such a manner that in the presence of voiced speech signals the low-pass filter with the higher limiting frequency and in the presence of unvoiced speech signals the low-pass filter with the lower limiting frequency is turned on.

This arrangement may be modified in a very advantageous manner by providing only one low-pass filter for each spectrum channel the upper limiting frequency of which is adjustable to a lower value by means of a controllable switch.

One modification of the method of improving the speech quality in connection with the analysis of unvoiced speech segments according to the channel vocoder principle consists in the provision that the noise component of the DC. voltage signal carried by each spectrum channel after the analysis is reduced by replacing the smoothing low-pass filter in each spectrum channel with an integrator which is controllable with respect to its integration time and which in the presence of unvoiced speech segments integrates the signals of the channel within an integration interval increased with respect to the normal integration interval.

An arrangement for carrying through this method gives especially favorable results due to the provision that for each spectrum channel there is provided an integrator following the DO. circuit and controllable with respect to its integration interval, the control input of which is connected through a switch to two control lines in such a manner that, by means of avoiced/ unvoiced discriminator known per se and included in the excitation channel, said switch in the presence of unvoiced speech segments connects said integrator to one signal generator, whereby the signals of the spectrum channel are integrated within a longer interval of time than in the presence of voiced speech segments when the integrator is connected to another signal generator and the signals of the spectrum channel are integrated within a normal interval of time.

By using the above indicated methods and arrangements, it is possible to improve the quality of the synthetic speech due to the fact that when analyzing unvoiced speech segments the integration intervals for the integration of the channel Voltages are made longer than it is e.g. necessary for an optimum integration of the channel voltages when analyzing voiced speech segments.

The invention will be described in detail below in conjunction with an embodiment thereof as illustrated in the drawing, wherein:

FIG. 1 shows the block circuit diagram of the analysis section of a channel vocoder,

FIG. 2 shows the block circuit diagram. of an arrangement for the excitation-controlled smoothing of the channel voltages by switching to two low-pass filters having different upper limiting frequencies,

FIG. 3 shows the basic circuit diagram of a low-pass filter having an adjustable upper limiting frequency,

FIGURE 4 shows the block circuit diagram of an arrangement for the excitation-controlled smoothing of the channel voltages by means of controllable integrators, and

FIG. 5 shows the basic circuit diagram of a controllable integrator.

In the ideal case, the analysis of voiced or unvoiced speech segments according to the channel vocoder principle results in constant output D.C. voltages unless the speech segments fall Ibelow a certain given length. This ideal case is easily obtainable for voiced speech segments.

For unvoiced speech segments, on the other hand, the output voltages contain noise components the energy spectrum of which corresponds to the integration interval of the integration member LP, following the DC. circuit G, (FIG. 1) in each spectrum channel. Also corresponding to the integration interval of such an integrator is the band width of a smoothing low-pass filter, so that for reducing the noise component it is necessary either to select a low value for the upper limiting frequency of the smoohting low-pass filter or to select a large value for the integration interval of the integrator used. However, a low upper limiting frequency of the smoothing lowpass filter or a large integration interval of an integrator result in a reduced flow of information, which also has an unfavorable effect on the quality of the synthetic speech. Forunvoiced segments of speech, this gain in quality is compensated by a reduction of the noise components. For voiced speech segments, the reduction of the limiting frequency below a certain value or the selec tion of a too large integration interval results in a loss of quality in the synthetic speech. Therefore, it is of advantage to select a different optimum limiting frequency of the smoothing low-pass filter or a different integration interval of the integrator for unvoiced segments of the speech signal than for voiced segments.

In most of the known vocoder systems, a discriminator D is used in the excitation channel, as shown in FIG. 1, for making a voiced/unvoiced discrimination sh and sl. This discrimination may be used for controlling either the limiting frequency of the smoothing low-pass filters or the integration interval when using integrators, so that for both casesvoiced and unvoiced speech segmentsoptimum values are set up automatically.

FIG. 2 shows the block circuit diagram of a channel vocoder using smoothing low-pass filters controllable with respect to their upper limiting frequency. In the ensuing discussion, the subscript i denotes any instant example of a part shown in the \drawing with the subscript 1 through n. In this arrangement, two lowpass filters LP, and LP, are provided for each spectrum channel which have differential upper limiting frequencies. A controllable switch S, is used for connecting, when analyzing unvoiced speech segments, the low-pass filter LP, with the lower upper limiting frequency through the contact A, into the spectrum channel K,, depending on the output signal of the voiced/unvoiced discriminator D. In order to avoid an unnecessary reduction of the flow of information in the presence of voiced speech segments, the low-pass filter LP, with the upper limiting frequency optimal for the analysis of voiced speech segments is connected into the channel by means of the controllable switch. It is also particularly advantageous to use a low-pass filter LP with a reversible upper limiting frequenoy, such as it is shown basically in FIG. 3. It consists essentially of two serially connected partial inductances L, and L, and a capacitor C, which by means of the switch S, may be connected either to the end of partial inductance L, or to the central tap between the two partial inductances. There, the relationship of L,, L, and C, is selected so that in the two switch positions A, and A,, for the analysis of voiced speech segments and for the analysis of unvoiced speech segments respectively, the optimum upper limiting frequency may be set up. When using this controllable low-pass filter LP, in the arrangement of FIG. 2, the pairs of low-pass filters LP, and LP, are replaced with these filters.

Another favorable possibility of realizing a low-pass filter with a controllable upper limiting frequency is obtained by using triggered integrators. Quite generally, an integrator may be imagined as a low-pass filter which in response to the Dirac impulse produces a square pulse of the duration of the integration interval and the output voltage of which is scanned at intervals each comprising one integration interval. Here, an increase in the integration interval corresponds to a reduction in the upper limiting frequency of this low-pass filter. This type of realization may be used to advantage wherever the channel volttages are to be represented digitally since the integrator already supplies a pulse-amplitude modulated signal. This signal then merely requires conversion in an analog-todigital converter.

FIG. 4 shows the Iblock circuit diagram of an arrangement in which triggered integrators are used for reducing the noise component of the spectrum signals when analyzing unvoiced speech signals. In this arrangement, depending on the output signal of the voiced/ unvoiced discriminator D, the triggered integrator I, is connected through the controllable switch S, either to a pulse generator supplying a signal I which consists of square pulses of a given pulse repetition rate which is selected optimally e.g. for the analysis. In the presence of unvoiced speech signals, the integrator has its trigger input a connected via the controllable switch to a generator supplying signals 1 which consist of square pulses with a pulse repetition rate lower than the pulse repetition rate of the signals t but selected optimally for the analysis of unvoiced speech signals.

FIG. 5 shows the basiccircuit diagram of such a triggered integrator. From the output of the DC. circuit G, of the respective channel K,, a DC. voltage of a constant amplitude or an A.C. voltage the envelope of which changes only very slowly is transmitted in the ideal case via line b to the input of the integrator I,. In order to illustrate the operation of the integrator circuit, let it be assumed that the capacitor C be charged negatively due to a preceding switching process at the transistor TR The voltage at the capacitor is assumed to be -V 0. As long as the transistor T R is cut off, the capacitor is discharged by the collector current i of the transistor TR,. The value of the resistor R is chosen so that the voltage droping thereacross is always low as compared to the voltage at the capacitor C. By the discharge, the voltage at the capacitor is continuously decreased. Wheri due to a positive pulse at the trigger input a of the generator t or t the transistor TR is momentarily turned on, the capacitor is simultaneously recharged to the voltage V The pulse-type charging current causes a pulse at resistor R the maximum amplitude of which is proportional to the voltage change at the capacitor. The signal of the shorttime mean value of the channel filter output voltage is derived at the electrode E and transmitted via the respective channel K',.

What is claimed is:

1. A method for improving the quality of speech signals analyzed in accordance with the channel vocoder principle wherein the output of each spectrum channel is a. signal representative of the energy of a corresponding band of frequencies in the input speech signal, which comprises smoothing said output signals to a greater degree for unvoiced than for voiced segments of the input speech signals,

by limiting the maximum rate of variation in said output signals to a lower value for said unvoiced than for voiced speech segments.

2. In a channel vocoder speech signal analyzing apparatus having speech signal input means,

band pass filter means connected to said input means and adapted to divide the input signal therefrom into a plurality of spectrum channels of different frequency bands,

D.C. circuit means connected to each filter means and adapted to detect the energy level of the signal in each spectrum channel,

and discriminator means connected to said input means and adapted to yield a signal indicative of the voiced/ unvoiced character of the input speech signal,

the improvement which comprises,

5 6 signal smoothing means connected to the output of each of said smoothing means comprises an adjustable said D.C. circuit means in each channel, low pass filter, and said smoothing means having a control input and lbeing said control input is operative to alter the cutoff freadjustable thereby between higher and lower upper quency of said filter. frequency limiting modes of operation, 5 5. Apparatus in accordance with claim 2, wherein, said control input being connected to said discriminator each of said smoothing means comprises an integrator means for operation of said smoothing means in having a variable integration interval.

the higher frequency mode during analysis of voiced segments of speech and in the lower mode during References Cited unvoiced segments of speech. 10 UNITED STATES PATENTS 3. Apparatus in accordance with claim 2, wherein, 2,810,787 10/1957 Di Toro et al. 179-15 each of said smoothing means comprises 3,261,916 7/1966 Bakis 179--1 two low pass filters of differing cutoff frequencies, and 3,102,928 9/ 1963 Schroeder 179-1 switch means under the control of said control input 3,012,098 12/1961 Riesz 179--1 to connect one or the other of said filters in circuit 5 in the Corresponding channeL KATHLEEN H. CLAFFY, Przmary Exammer.

4. Apparatus in accordance with claim 2, wherein, AY OR. A istant Examiner.