EP0094681B1 - Arrangement for electronic speech synthesis - Google Patents

Abstract

A circuit for electronically synthesizing speech has an audio generator for representing voiced sounds and a noise generator for representing voiceless sounds and a means for selecting significant parameters of the various speech elements by sampling and a means for storing those parameters. The circuit also includes a filter unit comprised of a number of individual filters and a means for selectively driving only those individual filters having filter coefficients necessary for representing the significant parameters of the particular speech element to be synthesized. The filters can be utilized individually or combined into selected groups in order to generate longer speech segments. The electronic signal at the output of the filter unit is edited for acoustically reproducing the desired speech elements and segments.

Description

The invention relates to a circuit arrangement for electronic speech synthesis, in which speech elements are represented by significant parameters and individual speech elements can be combined to form longer speech segments, in which an excitation signal for representing voiced and unvoiced sounds by a pulse generator and a noise generator using at least some of the significant ones Parameters are generated and fed to a filter circuit and in which the electrical signals on the filter output side are used for the acoustic reproduction of the desired speech elements and segments, the filter circuit consisting of individual filters arranged in parallel with fixed filter coefficients.

Methods for generating language elements are already known which can be put together to form longer language segments; these methods can be divided into the following two groups. The first group includes the methods in which the speech elements which were initially scanned in an analysis, possibly digitized and stored, for example, in a read-only memory, are retrieved from the memory and put together again for speech synthesis; redundant components that are not necessary for the intelligibility of the language are also stored, so that in this way good quality speech can be generated per se, although a correspondingly high memory requirement is required to display an extensive vocabulary. In the second group of speech synthesis methods, redundant speech components are largely hidden and the speech is stored in the form of significant parameters of speech elements, from which speech that is easy to understand for the listener can be generated with a significantly lower memory requirement. The heart of known circuit arrangements for performing the latter method are filter circuits with variable filter coefficients. For example, a speech synthesis circuit is known from DE-AS2209548, in which an excitation signal to which significant speech parameters are applied is fed to a filter circuit with variable filter coefficients. These filter coefficients are continuously controlled by further significant speech parameters during the entire synthesis process, so that the circuit arrangement for carrying out these methods must have devices for storing these filter coefficients. In addition, the circuit arrangement is to be provided with control devices for retrieving these coefficients from the memory and feeding them into the filters. Such tunable filters have relatively large dimensions and are difficult and difficult to implement with the narrow tolerances required for good speech quality.

From the "Journal of the Acoustical Society of America", Vol. 66, No. 5, November 1979, page 1328, a vocoder is known which consists of a plurality of individual filters arranged in parallel, which can be connected on the input side to a tone or noise generator. To generate speech, all individual filters are activated at the same time. All filter output signals are multiplied by continuously calculated spectral amplitude values and all individual values obtained in this way are added up.

The invention is based on the object of providing a circuit arrangement for electronic speech synthesis which is easier to implement and which generates easily understandable speech.

The object is achieved according to the invention in that the filter circuit is connected to a filter alternating clock generator which is used for the sequential control of individual specific individual filters and has a pulse period between 10 and 24 ms.

The invention has the advantage of a lower memory requirement, combined with the advantage that filters with fixed coefficients can be implemented more easily.

Further advantageous refinements of the inventions result from the subclaims. Thus, individual filters constructed in analog technology can be provided which can be acted upon by a time-discrete analog excitation signal; the implementation of such analog filters is particularly simple. This applies all the more if they are expediently constructed in a further embodiment of the invention as a transversal filter using so-called CCD technology. Alternatively, individual filters constructed in digital technology can also be provided, to which a time-discrete digital excitation signal can be applied, which has the advantage of being able to store the parameter values of the speech signals in a particularly simple manner.

In a further embodiment of the invention, the individual filters can be addressed individually for the display of speech elements. A circuit arrangement according to the invention can contain such a number of individual filters that all phonemes of a certain language can be represented. Several phonemes can be generated in a specific chronological order and connected to one another according to the characteristics of the human voice.

In a further embodiment of the invention, the individual filters can be interconnected in filter sets to represent longer language segments, the filter sets being controlled by addressing specific to the filter set. Such an alternative is characterized by lower memory requirements and is particularly suitable for the representation of speech in which the same speech segments occur repeatedly. The individual filters can also be arranged in a matrix, in each of which the individual filters of a matrix row are parallel to the respective address supply signal are applied, and the individual filter outputs of one matrix line can be connected sequentially to the matrix output. In a further embodiment of the invention, the individual filters can be designed as linear prediction or formant filters. The formant filters have fixed formant center frequencies and bandwidths. Language elements are represented by rendering at least the three lowest formants.

In a further embodiment of the invention, the individual filters can be implemented using what is known as CCD technology. They can also be designed as a transversal filter.

• FIG. 1 ein Blockschaltbild einer Schaltungsanordnung zur Sprachsynthese.
• FIG. 2 zeigt das Blockschaltbild eines Ausführungsbeispiels der Erfindung mit in Matrixform angeordneten Einzelfiltern.
• FIG. 3 und FIG. 4 zeigen die Verwendung von Linearprädiktions- bzw. Formantfiltern.

The invention will now be described in more detail to the extent necessary for understanding with reference to drawings. It shows

• FIG. 1 shows a block diagram of a circuit arrangement for speech synthesis.
• FIG. 2 shows the block diagram of an embodiment of the invention with individual filters arranged in matrix form.
• FIG. 3 and FIG. 4 show the use of linear prediction or formant filters.

FIG. 1 shows a block diagram of a circuit arrangement according to the invention, which in its core comprises a filter circuit F, an excitation generator device G and a control unit StE, which in turn are connected to an input unit EG or a low-pass filter TP and an electroacoustic transducer.

The input unit EG may output information about the speech elements to be synthesized to the control unit StE. This information can be entered, for example, using a keypad. Likewise, information about language elements to be synthesized can also be supplied from external systems in a system-conforming representation. The control unit StE may have facilities for temporarily storing and processing this information, e.g. in the so-called "handshake mode" with the filter circuit F and include memories in which speech parameters are stored. As shown in FIG. 1 is indicated, the control unit StE can lead to the filter circuit F, for example, two line connections, namely one for a filter change clock signal Tw and one for a digital speech element selection signal SEA, the filter change clock signal in the filter circuit F controlling the synthesis of the speech element that is generated by the speech element selection signal SEA is determined. The filter circuit F, which can be designed in the manner shown in FIG. 2, for example, contains, among other things, individual filters (F11 ... in FIG. 2) with fixed coefficients. With the aid of these individual filters, the speech synthesis is carried out, the individual filters supplying an output SA of the filter circuit F with an (electrical) speech signal which, if necessary after digital / analog conversion, is fed via a low-pass filter TP and, if appropriate, via a subsequent amplifier to an electroacoustic converter. As shown in FIG. 1 is further indicated, the filter circuit F can supply the control unit StE via a control line E with a digital signal which indicates the end of the synthesis process of a speech element and, in the handshake operation already mentioned, requests the information required for the synthesis process of the subsequent speech element determined by the information entered .

FIG. 2 shows circuit-specific details of an exemplary embodiment of the filter circuit F, the excitation generator device G and the control unit StE.

The representation in FIG. 2 from an arrangement of the individual filters in a matrix. The filter circuit F includes the columns F11, F21, ... Fnl; ... and lines F11, F12, ... F1Z; ... arranged individual filters, line-specific multiplexers M1, M2, ... Mn a line selection multiplexer ZMF, a selection circuit ZME and, if appropriate, one in FIG. 2, not shown in detail, inserted between the excitation generator device G and the individual filters.

According to FIG. 2, the excitation generator device G consists of a controllable pulse and noise generator IG or RG and a switching device. The control unit StE includes, among other things, memories S1 ... Sn, in which speech parameter values are stored, a filter change clock generator FwG, and a memory selection circuit ZMA. The filter circuit F receives a filter change clock pulse Tw and a speech element selection signal SEA from the control unit StE. The filter change clock generator FwG arranged in the control unit StE generates equidistant filter change clock pulses with a pulse period Tw, which can be, for example, between 10 and 25 ms. The filter change clock pulses are simultaneously fed to all line-specific multiplexers M1 ... in the filter circuit F and the memories S1 ... arranged in the control unit StE. In the exemplary embodiment considered here, the number of line-specific multiplexers M is equal to the number of said memories S. This number corresponds to the number of lines in the filter matrix. If - as assumed above - different speech segments can be generated by the filter circuit Fn, the filter circuit Fn has filter sets arranged in rows. Each filter set contains at least one individual filter. The language segment generated in the filter set can be composed of several language elements that are generated in the filters belonging to this filter set. The duration of a language element is t _w ; the duration of a language segment composed of m language elements is then mt _w . The number of individual filters required for the generation of such a language segment can be less than m if the language segment in question contains the same language elements that are synthesized in the same individual filters of the filter set in question. The (analog) language elements Signals are interconnected by the respective line-specific multiplexer M in the filter change cycle Tw to form the (analog) speech signal SA. The pulse sequence generated by the filter change clock generator FWG with the frequency 1 / t _w is also supplied to all line-specific memories S in the control unit StE, in which the parameter values of the excitation signals, e.g. B. their frequency f and amplitude U are stored. Due to the pulse sequence generated by the filter change clock generator FwG, these parameter values are retrieved from the memories S and fed to the memory selection circuit ZMA. The latter then selects the parameter values of the speech segment to be generated in accordance with the speech element selection signal SEA which is also supplied to it and feeds them to the controllable excitation generator device G. The excitation generator device G comprises a pulse generator IG which can be controlled in frequency and amplitude, and a noise generator RG which can be controlled in amplitude. The switching device provided on the output side in the excitation generator device G is controlled by the information about the frequency retrieved from the memories S: for frequency values equal to zero, the noise generator RG is coupled to the filter circuit F, for frequency values not equal to zero the pulse generator IG is connected to the filter circuit F coupled. Depending on the parameter values f and U, the excitation generator device G delivers pulse or noise signals of a certain amplitude and possibly frequency. Unvoiced speech elements are simulated by noise signals, voiced speech elements of a certain frequency by pulse trains of precisely this frequency. In the exemplary embodiment described here, the excitation signals generated by the controllable excitation generator device G are supplied to all filter sets, including those that are not used to generate the selected speech segment. All (analog) signals generated in the filter sets are fed via the line-specific multiplexers M to the line selection multiplexer ZMF, in which the desired speech signal is selected by means of the speech element selection signal SEA, which then appears at the output SA of the line selection multiplexer. The voice signal SA is fed to the low-pass filter TP, which filters out higher frequency components contained in the voice signal, for example on the basis of pulse-like excitation of the filters. It should be noted here that the above explanations are not limited to a speech signal synthesis by means of analog filters loaded with analog excitation signals, but also apply in a corresponding manner to a speech signal arrangement by means of digital filters loaded with digital excitation signals, the filter output signal then also being a digital-analog signal. Undergoes conversion. After any amplification that may still be required, the electrical analog voice signal is finally reproduced via an electroacoustic transducer.

The line-specific multiplexers M emit a digital signal E to the selection circuit ZME simultaneously with the switching through of the last speech element generated in the filter set in question, which characterizes the time completion of the speech synthesis process in the filter set. Like the line selection multiplexer ZMF, the selection circuit ZME brought from the current speech element selection signal SEA into a corresponding switching position now switches the corresponding digital signal through to the control unit StE, which can thus trigger the synthesis process of the next speech segment.

Deviating from the representation in FIG. 2, the individual filters with fixed coefficients can also be addressed individually. In this case, the filter circuit (F in FIG. 2) has no line selection multiplexer (ZMF in FIG. 2) and no selection circuit (ZME in FIG. 2) and no line-specific multiplexer (M in FIG. 2) of the type described above. Rather, the control unit (StE in FIG. 2) then has devices for storing individually addressable parameter values for the excitation signals and for interconnecting the speech element signals that can be generated in the individual filters. The filter change clock generator FwG (in FIG. 2) and the controllable excitation generator device G (in FIG. 2) (possibly with a time window circuit) also perform the functions described above in this embodiment. Such an embodiment, characterized by optional control of individual filters by means of individual filter-individual addressing, need only have different individual filters, while in the exemplary embodiment according to FIG. 2 the same individual filters can also be arranged in the different filter sets, possibly also in one and the same filter set. The latter alternatives, which can be implemented with less control effort due to the filtering-specific addressing compared to the single-filtering addressing, are particularly suitable for the reproduction of speech that repeatedly contains the same language segments. Embodiments that contain both individual filters connected in filter sets and independent individual filters will be of practical importance. In this way, the number of individual filters used and the necessary tax expenditure can be optimized.

The individual filters used according to the invention can also be used as linear prediction filters with fixed coefficients according to FIG. 3 be formed. Linear prediction as such is known and described in the relevant specialist literature (such as FLANAGAN, Speech Analysis Synthesis and Perception, Springer-Verlag Berlin, Heidelberg, New York 1972, p. 367 ff., P. 390 ff.), So that a detailed description is unnecessary here. The achievable speech quality is proportional to the number of coefficients within certain limits. Good speech quality can already be achieved with around 10 filter coefficients. A sol The prediction filter is shown in FIG. 3 connected with A11 and B11 to the correspondingly designated connection terminals of the filter circuit F according to FIG. 2 2, in order to form the individual filter F11 there. The linear prediction filters used according to the invention can be designed as analog or digital filters. Accordingly, the filters are supplied with excitation signals in analog or digital form from the excitation generator; accordingly, analog or digital signals result at the filter outputs.

The individual filters used according to the invention can, as shown in FIG. 4, also be designed as formant filters with fixed filter coefficients, wherein each individual filter can correspond to a parallel connection of three formant filters for emulating at least the first three (low-frequency) speech formants. The generation of speech by formant synthesis is known and sufficiently described (for example in the literature FLANAGAN p. 339 et seq. Already cited), so that there is no need to describe it here. The formant filters are expediently in the form of bandpasses with certain fixed pass bands and center frequencies of these pass bands. Such a filter circuit can also be implemented in analog and digital technology.

The individual filters can be implemented in so-called CCD technology in all of the above cases. When using transversal filters or recursive filters, the excitation signal is fed to the individual filters in a time-discrete form. For this purpose, the filter circuit (F in FIG. 2) can be shown in FIG. 2 time slot circuit not shown included. According to the sampling theorem, the time window circuit can generate a sampling signal of fixed frequency, which has at least twice the frequency with respect to the network signal to be sampled. The controllable excitation generator device G and all individual filters are clocked into the filter circuit F with this scanning signal generated in this way.

Claims (11)

1. A circuit arrangement for electronic speech synthesis, wherein the speech elements are represented by significant parameters and individual speech elements can be assembled to form longer speech segments (SA), wherein an excitation signal for the representation of vocal and non-vocal sounds is produced by a pulse generator (IG) and a noise generator (RG) by employing at least a portion of the significant parameters and is supplied to a filter circuit (F) and wherein the electrical signals at the filter output end are used for the acoustic reproduction of the required speech elements and speech segments, where the filter circuit (F) comprises parallel individual filters (F11, F12,..., F1M1; F21, F22,...; Fn1 ...) and has a pulse period of between 10 and 25ms.

2. A circuit arrangement as claimed in claim 1, characterised in that there are provided individual filters which are constructed in analogue technology and can be acted upon by a time-discrete analogue excitation signal.

3. A circuit arrangement as claimed in claim 1, characterised in that there are provided individual filters which are constructed in digital technology and which can be acted upon by a time-discrete digital excitation signal, and that the obtained speech signal is digital/analogue-converted for the speech reproduction.

4. A circuit arrangement as claimed in claim 2 or 3, characterised by the random control of individual filters by individually addressing individual filters.

5. A circuit arrangement as claimed in claim 2 or 3, characterised in that the individual filters are arranged in filter sets for the representation of long speech segments, where a random control of the filter sets is effected by filter set-individual addressing.

6. A circuit arrangement as claimed in claim 5, characterised in that the individual filters are arranged in a matrix, wherein the individual filters of a matrix line can be respectively acted upon in parallel by the respective excitation signal and the outputs of the individual filters of one matrix line can be sequentially connected to the matrix output.

7. A circuit arrangement as claimed in claim 5 or 6, characterised in that for each filter set a storage unit is provided wherein parameters of the excitation signal for the representation of a speech segment are stored.

8. A circuit arrangement as claimed in one of claims 4 to 7, characterised in that the individual filters are designed as so-called linear predictive filters.

9. A circuit arrangement as claimed in one of claims 4 to 7, characterised in that the individual filters are designed as formant filters having fixed formant centre frequency coefficients and formant bandwidth coefficients for the production of speech by the representation of at least the three lowest formants.

10. A circuit arrangement as claimed in one of claims 4 to 9, characterised in that the individual filters are realised in so-called CCD-technology.

11. A circuit arrangement as claimed in one of claims 4 to 10, characterised in that the individual filters are designed as transversal filters.

Images