Combination speech synthesis and recognition apparatus

• US 4,520,499 A
• Filed: 06/25/1982
• Issued: 05/28/1985
• Est. Priority Date: 06/25/1982
• Status: Expired due to Fees

• Current Assignee: Milton Bradley, Co. (Hasbro Inc.)
• Original Assignee: Milton Bradley, Co. (Hasbro Inc.)
• Inventors: Vetter, David C., Skoge, Klaus, Ahrens, Paul R., Montlick, Terry
• Primary Examiner(s): Kemeny, E. S. Matt
• Application Number:
  US06/392,309
• Time in Patent Office: 1,068 Days
• Field of Search: 381/36-53, 381/513.5, 364/724
• US Class Current: 704/266
• CPC Class Codes: G10L 13/047 Architecture of speech synthesisers
• CPC Class Codes: G10L 15/28 Constructional details of speech recognition systems

Abstract

A reconfigurable lattice filter is employed to permit the same circuitry to function as a speech synthesizer and as a speech analyzer or recognizer. The lattice filter can be configured both as an all-pole filter (for synthesis) and as an all-zero filter (for analysis), with the choice being determined by the state of an analysis/synthesis signal (i.e., mode control signal) provided thereto. The connections between various elements in the circuitry are controlled by the analysis/synthesis signal, also.

In synthesis mode, partial correlation coefficients are supplied to the filter from a microprocessor. The filter is excited by a one of a number of stored patterns simulating a glottal pulse for voiced sounds and by a pseudo-random noise generator for unvoiced sounds.

In analysis mode, appropriate feedback control paths are enabled so as to provide to the filter coefficients which change in response to changes in the input speech waveform. Coefficient values thus determined are averaged over fixed intervals and successions of such averaged coefficient sets produce representations of words or phrases which can then be used for speech recognition.

Claims

1. An apparatus for synthesizing speech from stored representations and for generating representations of speech for use in speech analysis, including but not limited to speech recognition, such apparatus being reconfigurable between the two functions and comprising:

A. a lattice filter including elements which are configurable selectively to provide an all-pole transfer function for synthesis and an all-zero transfer function for analysis, the lattice filter being responsive to coefficients supplied thereto;

B. an input converter for converting analog representations of acoustic waveforms to signals suitable for processing by said all-zero lattice filter;

C. an output converter coupled to said all-pole reconfigured lattice filter for coverting the output signals from said filter to analog signals suitable for conversion into acoustic waveforms;

D. an excitation source for supplying to the input of said all-pole filter signals representing vocal tract excitation;

E. a filter coefficient correction means for modifying lattice filter coefficients responsive to a speech signal being analyzed, to generate representations of such speech;

F. means for receiving filter coefficients for speech synthesis;

G. means for selectively interconnecting, for speech synthesis or analysis, the lattice filter, the input converter, the output converter, the excitation source, the filter coefficient correction means and the coefficient receiving means.

5. An apparatus as defined in claims 1, 2, 3 or 4 wherein said input converter means includes a means for normalizing the amplitude of the input to said lattice filter, whereby the time required to achieve a set of filter coefficients representing the input acoustic waveform is substantially independent of the amplitude of the input acoustic waveform.

6. An apparatus as defined in claim 5 in which the amplitude normalizing means include:

A. a variable gain circuit;

B. means for controlling the gain of the variable gain circuit in accordance with the average amplitude of the input acoustic waveform over a predetermined length of time; and

C. means for effecting a reduction in the gain of the variable gain circuit in response to an instantaneous acoustic waveform amplitude in excess of a threshold corresponding to the gain of the variable gain circuit immediately before the gain reduction.

13. An apparatus as defined in claim 1 including

A. a first controllable scaler for controlling the amplitude of the excitation applied to said lattice filter when the filter is in its all-pole configuration;

and

B. a second controllable scaler for adjusting the amplitude of the output of said lattice filter, whereby the maximum amplitude of the signal passing through the filter may be maintained substantially at the maximum amplitude of which the filter is capable, thereby to minimize noise, and the amplitude of the output from the output converter may be adjusted to generate the overall energy variations in the speech signal being synthesized.

2. An apparatus for synthesizing speech from stored representations and for generating representations of speech for use in speech analysis, including but not limited to speech recognition, such apparatus being reconfigurable between the two functions and comprising:

A. a lattice filter selectively configurable to have an all-pole transfer function for synthesis and an all-zero transfer function for analysis, the lattice filter being responsive to coefficients supplied thereto;

B. an input inverter for converting analog representations of acoustic waveforms to signals suitable for processing by said all-zero lattice filter;

C. an output converter coupled to said all-pole reconfigured lattice filter for coverting the output signals from said filter to analog signals suitable for conversion into acoustic waveforms;

D. an excitation source for supplying to the input of said all-pole filter signals representing vocal tract excitation;

E. a filter coefficient correction means for modifying lattice filter coefficients responsive to a speech signal being analyzed, to generate representations of such speech, and comprising

1. a feedback control means for adapting the response of said lattice filter to signals from said input converter, and

2. a coefficient integrator and storage means coupled to said feedback control means for integrating the output of said feedback control means and storing the results of the integration, said results being a set of coefficients characteristic of the input acoustic waveform;

F. means for receiving filter coefficients for speech synthesis;

G. means for selectively interconnecting, for speech synthesis or analysis, the lattice filter, the input converter, the output converter, the excitation source, the filter coefficient correction means and the coefficient receiving means.

3. An apparatus as defined in claim 2, further comprising a coefficient averager and buffer coupled to said coefficient integrator and storage means for time-averaging each of the said coefficients.

4. An apparatus as defined in claim 3 wherein the feedback control means includes a means for multiplying the scaled forward prediction error from each stage of said lattice filter by the sign of the backward prediction error from that same stage.

7. An apparatus as defined in claim 2 including means for configuring the coefficient integrator and storage means, when the lattice filter is in its all-pole configuration, as a working storage that holds the filter coefficients and applies them to the filter during speech synthesis.

8. An apparatus as defined in claim 3 including means for configuring the coefficient integrator and storage means, when the lattice filter is in its all-pole configuration, as working storage that holds the filter coefficients and applies them to the filter during speech synthesis.

9. An apparatus as defined in claim 8 including means for configuring the coefficient averager and buffer, when the lattice filter is in its all-pole configuration, to serve as a buffer for temporarily storing filter coefficients received by said receiving means and then transferring the coefficients to said working storage.

10. An apparatus as defined in claim 2 in which said coefficient integrator and storage means includes:

A. a shift register having an input end and an output end;

B. an adder having first and second input terminals and providing an output that is the sum of the inputs applied to its input terminals;

C. means applying the adder output to the input end of the shift register;

D. a first gate for selectively connecting the output end of said shift register to the first input terminal of the adder, and

E. a second gate for selectively applying an input to the second input terminal of the adder.

11. An apparatus as defined in claim 10:

A. in which said feedback control means develops a scaled value of the forward prediction error from said lattice filter; and

B. including means for selectively applying said scaled value to said second gate when said lattice filter is in its all-zero configuration.

12. An apparatus as defined in claim 11 in which:

A. said adder responds to a sign signal by adding its input signals when the sign signal has a first state and subtracting one of its inputs from the other when the sign signal has a second state; and

B. said control means includes means for developing the sign signal in response to the sign of the backward prediction error from said lattice filter.

14. An apparatus as defined in claim 10:

A. in which said shift register contains a plurality of fields holding filter coefficients and at least one field holding an elapsed time;

B. including means for applying a fixed number to said second adder input terminal when said time field is applied to said first adder input, thereby to change said elapsed time by a fixed amount on each recirculation of said shift register, and

C. including means for detecting a predetermined value of said elapsed time.