DE1258472B - Arrangement for changing the basic tone, the speaking speed and the timbre of speech signals analyzed according to the vocoder principle - Google Patents

Description

Arrangement for changing the basic tone, the speed of speech and the timbre of speech signals analyzed according to the vocoder principle. The invention relates to an arrangement for the optional change of the fundamental tone, the speed of speech and the timbre of analyzed according to the principle of the pulse-excited channel vocoder Speech signals. In the case of the channel vocoder with impulse excitation, the Input speech signal has two functions, namely a spectrum function and an excitation function derived.

In order to vary the fundamental tone while maintaining the same timbre of speech, it is necessary to determine the fundamental frequency of the excitation including its harmonics to vary in frequency, and it is also necessary to also use the spectrum values to shift to lower or higher frequencies.

A mere change in the language spectrum causes above all one Change in the sound impression of the language, which changes while maintaining the original Represents fundamental frequency as a different timbre.

If only the basic frequency is changed, there is also one changed timbre, since now the original ratio of the fundamental frequency to their harmonics is disturbed.

Similar problems arise when changing the articulation speed, which, if done alone, also changes the timbre and the pitch evokes. Should the articulation or speaking speed be without these side effects can be changed, special measures must be taken will. The object of the invention is to provide an arrangement with the help of which it is possible, optionally the height of the fundamental tone, the speaking speed and to vary the timbre of the language.

For an arrangement for the optional change of the fundamental tone, the Speech speed and the timbre of according to the principle of the impulse-excited Channel vocoders analyzed speech signals, the invention is that to change of the fundamental tone in the excitation channel a modifying circuit is switched on, which is the digital value of the distance between two consecutive zero crossings a direction of the speech fundamental wave characterizing impulses (excitation impulses) numerically changed that means for changing the speed of speech in the excitation channel are provided, each of the series of incoming excitation pulses optionally n-th does not transmit or every n-th transmits several times in an interval, and that for Change of timbre means are provided, which are in digatal form present samples of the spectrum channels for each channel as a function of change its neighboring channels numerically.

The language changes mentioned in the previous paragraph can can be made both on its own and in combination, with the combination will always be chosen if z. B. the fundamental without changing the timbre or the speed of speech without changing the fundamental tone and timbre should be changed.

The invention is applicable wherever with good results for example, data from electronic data processing systems in spoken form Form are to be output, such as B. Information on stock exchange prices or seat reservations, and where data about the language, e.g. B. the values of the momentary energy in the individual Spectrum channels or the information about the excitation function, in digital form z. B. as pulse code groups or as groups of pulse amplitude modulated signals are stored. Because of this variation, especially the pitch of the root and the Speech speed, can be achieved once that the transmission of the spoken Output data (change in the frequency of the fundamental tone) better to a transmission channel can be adjusted and on the other hand (change in the rate of speech ), .that the output speed of the spoken data is influenced.

The invention is illustrated by means of an exemplary embodiment illustrated by drawings described in more detail. It shows F i g. 1 shows the block diagram of an arrangement for changing the basic tone, the timbre and the speaking speed of speech signals, F i g. 2 a representation of the time shift of the energy values in the individual Spectrum channels, FIG. 3 shows an analog representation of the channel values of the spectrum channels at various points in the arrangement and FIG. 4 the block diagram of an extended Excitation channel.

The arrangement according to FIG. 1 analyzes the spectrum of the speech signal s (t) in a known manner through the individual frequency bands matched bandpass filters BP to BP, which are preferably designed logarithmically so that each bandpass filter allows an equal interval in the spectrum to pass through, as well as downstream channels K1 to K5 Rectifiers G1 to G5 and low-pass filters LP to LP .., which determine the instantaneous value of the energies in the individual frequency bands, the so-called channel values. These channel values A to E are, as in FIG. 1, scanned by a switch AS, and the analog quantities of the channel values are converted into digital quantities, preferably in binary parallel representation, in an analog-to-digital converter ADW. The values present in parallel at the output of the converter ADW are then converted into series values by a further scanner PSWi.

Since the individual channel values in this series display do not carry an identifier about the channel from which they were derived, the assignment of a channel value to its frequency band is only possible through the occurrence of these values over time. In the one shown in FIG. 1 and 2, the five channels K1 to K5 are scanned one after the other at time intervals k1 to k and the values A to E are derived. In. the individual lines of FIG. 2 are the channel values A to E at the correspondingly designated circuit points of the arrangement according to FIG. 1 entered as a function of time t. The value which occurs in the time interval Ici at point (e) is interpreted by the following devices as the value of the frequency band of channel K1. In the arrangement shown, it is assumed that a channel value is represented by a four-bit long code group and is delayed by four bits by a delay element DL (4T) 1. The illustrated arrangement of two delay elements DL (4 T) 1 and DL (4 T) 2 with the switch S allows for the output (d) or (eN) to (eT) [Fig. 3] a shift of the channel values by one channel to the left or right, ie to lower or higher frequencies, namely in switch position H to higher and in switch position T to lower frequencies, while the middle switch position N corresponds to the normal position. If a shift should take place by several channels, correspondingly more delay elements DL (4 T) i would have to be provided in an analogous manner. The arrangement shown after the switch S allows a channel value to be influenced as a function of the adjacent channel values. The adjacent channel values of the channel value at the tap (e) between the delay elements DL (4 T) and DL (4 11) taken in points (d) and (f) are converted into Mult (n) 1, Mult (n) 2 in the circuits multiplied by a factor n <1 and added in the circuit Addi. This sum value (f.,) is then either added to the channel value or subtracted from the channel value (Add2 or Subi). The resulting value, ie the sum or difference, is also added with multiplied by a compensation factor F1 or F2.

The effect of this arrangement is better from FIG. 3 can be seen. For a better overview, the channel values A to E are represented here by analog values. Diagram 1 shows the original channel values A, B, C, D and E which occur in chronological succession at the tap (e) of the delay chain consisting of individual delay elements DL (4 T) i.

The diagrams shown under 2 and 3 show the values occurring in chronological order at taps (d) and (f) and multiplied by the factor n = 0.25.

Diagram 3 a then shows the values (f ") after the addition in the Addi circuit. The resulting sum of the modified adjacent channel values of the channel value occurring at tap (e) is either added to this (Add2) or subtracted from it (Subi The two alternative values are then still with a compensation factor multiplied and made available at the outputs (i) or (4 for further processing.

This numerical change in the spectrum is shown in Diagrams 4 and 5 of FIG. 3, where in FIG. 4 the subtraction of the values from 1 and 3 a with subsequent multiplication by the factor F., = 2 and in 5 the addition of the values of 1 and 3 a with subsequent multiplication by the factor Fi = 3 is. The marginal values neglected because of their small size, e.g. B. the first and last value of 3 a, are not used for further processing of the voice signals more considered. One sees, as in FIG. 4, a relative increase or in FIG. 5 a relative one Lowering of the peak values is achieved with approximately the same spectrum area. The multiplication of the channel values is an easy process. For the selected binary channel value representation and the factor n = 0.25, it is only one binary shift of the values by 2 T with trailing binary zeros for Mult (ra) 1.2. In an analogous way, the factors F1 and F., depending on the delay and addition depending on the choice of the values for n can be exactly represented or approximated.

The second function derived from the speech signal s (t) is the excitation function Ko, which is derived in a known manner (see FIG. 1) from the zero crossings of the fundamental speech wave in the circuit ND and consists of appropriately shaped pulses in the circuit PF whose repetition frequency is equal to the basic speech frequency. A counter Z, which is switched on by a fixed clock generator TG, converts the intervals between successive pulses into binary numbers, which are converted into series representation in a parallel-to-series converter PSW2. Up to this point the arrangement of the arrival corresponding to the reg ungskanals already become known prior art. If a resulting excitation is now to be used to generate a speech signal with a longer or shorter speaking duration without influencing the basic pitch, additional excitation values must be generated or derived values must be eliminated. If the fundamental tone is to be changed, the measured values of the excitation pulses must be modified. This change in the fundamental tone is on the one hand through numerical multiplication in a multiplier Mult (n) 3 in FIG. 1 or, on the other hand, by varying the frequency of the clock generator TG (VTG in FIG. 4). The change in articulation speed resulting in both cases (the analogue would be a faster or slower running tape) must be compensated for by inserting or leaving excitation pulses.

The F i g. 1 shows in block form an arrangement for changing the rate of speech. The circuit labeled PZC can selectively and at regular intervals pulses suppress or repeat from the circuit PF. The circuit labeled AT ensures the timed delivery of the impulses.

One possible embodiment for the .Schaltung PZC of FIG. 1 is in Fig. 4 included.

The pulse series emitted by the pulse shaper PF is reduced by one pulse by the rotary switch Si with each revolution, for example by separating a contact from the contact bank, and by the rotary switch S2 by doubling a pulse, for example by the delay in the circuit DLY2 and the OR circuit O , increased by one pulse. The switches themselves are switched one step further by each pulse of the pulse series via a rotary switch control DS connected to the pulse generator PF. Each pulse that reaches the sampling control ATS via the switch% with the switch positions T, N, H triggers a single sampling of the preferably binary value in the counter Z. The considered pulse of the series, which is delayed in DLYI, then resets the counter to 0 via its input Y, and a new counting process begins. The counter values are numerically influenced by changing the clock generator VTG that controls the counter. The parallel display of the binary value of the counter Z is converted into a series display by the switch PSW3 and transferred to the memory SP.

The described arrangement for analyzing and modifying the spectrum and excitation function supplies digital signals for the spectrum code and the excitation code, which represent the modified speech signal , at its outputs (e), (i), (l) and CA (FIG. 1) . If the modification does not include a change in the articulation speed, speech can be generated directly from the code. If the articulation speed is modified, however, a digital memory SP (FIG. 4) must be used which is able to buffer the difference in the data flow of the code generation and the speech signal regeneration resulting from the different time scales.

Claims (4)

Claims: 1. Arrangement for the optional change of the fundamental tone, the speaking speed and the timbre of speech signals analyzed according to the principle of the pulse-excited channel vocoder, characterized in that a modifying circuit [ Mult (n) 3] is switched on, which numerically changes the digital value of the distance between two consecutive zero crossings of a direction of the speech fundamental wave characterizing pulses (excitation pulses) that means (PZC) are provided for changing the speech speed in the excitation channel, which are from the series of incoming excitation pulses either every h-th not or every n-th in an interval several times, and that means [DL (4T) 1 to DL (4T) 4, S, Mult (n) l, "Addl, 2, Subl, F1 and F2] are provided which contain the samples of the spectrum channels (K1 to K5) for each Ka nal numerically depending on its neighboring channels (Fig. 1 and 3). 2. Modification of the arrangement according to claim 1, characterized in that that to change the fundamental tone in the excitation channel (K, in Fig. 4) in known Way provided digital counter (Z) from a pulse generator (VTG) with optional variable pulse repetition rate is controlled, increasing the digital value of the speech stimulus is modified numerically. 3. Arrangement according to claim 1 or 2, characterized in that to change the speech rate in the excitation channel (K, in F i g. 4) in connection with a pulse shaper (PF) a delay circuit (DLYI) follows, the output of which is connected to the zero input ( r) a counter (Z) is connected that a rotary switch control (DS) is provided, which is connected to the pulse shaper and advances two rotary switches (Sf, S2) by one step with each pulse, the contact banks of which are also connected to the pulse shaper in this way that in one switch (S ) a contact in its wiring to the contact bank is interrupted and in the other (S2) an additional contact is provided which leads to the output of an OR circuit (O), one input of which with the contact bank and the other input of which leads to a further delay circuit (DLY2), the input of which is also connected to the contact bank, that a further switch (S5) is also provided, de ssen first contact at the output of a rotary switch (Sf), whose second contact is on the contact bank and whose third contact is at the output of the other rotary switch (S2), and that the output of the further switch (S5) with a scanning control (ATS) for Control of a parallel-serial converter (PSW3) following the counter (Z) is connected. 4. Arrangement according to claim 1, characterized in that for changing the timbre of an analog-to-digital converter (ADW in F i g. 1) provided in the spectrum channel is followed by a parallel-to-serial converter (PSWl), the output of which with one of individual Delay lines [DL (4T) 1 to DL (4T) 4] with intermediate taps [(b), (c), (d), (e)] provided delay chain is connected, whose input [(a)] and whose first and second tap [(b), (c)] are placed on the contacts (T, N, H) of a selector switch (S), the output of which is connected to the third tap [(d)], which in turn becomes a multiplier circuit [Mult (n) 2] leads to the fact that the fourth tap [(e)] is connected both to the first input of an adding circuit (Add2) and to the first input of a subtracting circuit (Sub1), the remaining input being the adding and the subtracting circuit is connected to the output of a further adding circuit (Addi), one input of which leads to the output ng of the already mentioned multiplier circuit [Malt (yz),] and its other input leads to the output of another multiplier circuit [Mult (n) 1], the input of which is connected to the output [(f)] of the delay chain, and that the output the former adding circuit (Add2) is interconnected with the input of a first correction stage (F) and the output of the subtraction circuit (Subi) with the input of a second correction stage (F2), so that the varied spectrum codes occur at the outputs of the correction stages.