Input device for computer speech recognition system

• US 4,461,024 A
• Filed: 12/01/1981
• Issued: 07/17/1984
• Est. Priority Date: 12/09/1980
• Status: Expired due to Fees

• Current Assignee: The secretary of state for defence in her Britannic Majesty's Government of the United Kingdom of Great Britain And Northern Ireland
• Original Assignee: Secretary of State For Industry In Her Britannic Majesty's Government of The United Kingdon of Great Britain and Northern Ireland
• Inventors: Rengger, Ralph E., Manning, David R.
• Primary Examiner(s): Kemeny, E. S. Matt
• Application Number:
  US06/326,444
• Time in Patent Office: 959 Days
• Field of Search: 381/36-50, 364/513.5
• US Class Current: 704/233
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence

Abstract

Systems for the recognition of speech by computer have usually required t the speech be of a strictly standarized nature, free of such features as regional accent, and unaccompanied by background noise. The problem has been to produce a system providing information which will allow accurate recognition in the presence of noise, and with non-standard speech sounds. The invention provides to a computer in a speech recognition system information in parallel streams on a number of factors, viz, the existence, in any signal which may contain speech sounds, of a fundamental repetitive structure and its periodicity; the existence of a high frequency component having a wide frequency band; the existence of a component having energy relatively stable with respect to time and characteristic of background noise; the peak frequency, peak amplitude and band width of sounds lacking a low frequency component; and the frequency of resonant content of any component having a fundamental repetitive structure and the level of correlation.

Claims

1. A speech sound input device for providing information to a computer in a system of computer recognition of speech comprising

(i) a voiced monitor which provides to the computer a signal indicating whether the sound input to the device has a fundamental repetitive structure, and signals indicating the periodicity of any such structure;

(ii) an unvoiced monitor which provides to the computer a signal indicating whether the sound input to the device has no low frequency component and has at the same time high frequency components in a wide range of frequency;

(iii) a background monitor which provides to the computer a signal indicating the presence of relatively stable energy in the sound input to the device, characteristic of background noise with an absence of speech, and means for tracking the level of such relatively stable energy;

(iv) a noise color analyzer which provides to the computer, in relation to input sounds having no low frequency component, signals indicating the peak frequency of sound input to the input device, signals indicating the peak amplitude of said sound input, and signals indicating the bandwidth of said sound input; and

(v) a resonant frequency analyzer which provides to the computer, in relation to input sounds having a fundamental repetitive structure, signals indicating the frequency of the resonant content of the input sound and signals indicating its level of correlation.

2. A device according to claim 1 in which the resonant frequency analyzer provides signals related to low, medium and high frequency bands of speech frequency.

3. A device according to claim 2 in which the low frequency band is from 350 Hz to 700 Hz, the medium frequency band from 700 Hz to 1400 Hz, and the high frequency band from 1400 Hz to 2800 Hz.

4. A device according to claim 2 in which the resonant frequency analyzer includes a band pass filter corresponding to each of said frequency bands;

corresponding to each filter a tapped delay line; and

associated with each said delay line multiplexing means arranged to autocorrelate the outputs from predetermined taps on said delay line with the input to the delay line.

5. A device according to claim 4 in which the taps on each delay line are arranged in two groups such that auto correlation of outputs to input for a first group corresponds to a delay of the order of 3/4 wave length for that filter band, and the second group corresponds to a delay of the order of 1/2 wave length for the same filter band.

6. A device according to claim 1 in which the noise color analyzer has a plurality of band pass filters covering the range from 1 KHz to 5 KHz, a clocked multiplexer which scans the outputs from said filters in a period of the order of 1 millisecond to provide a frequency spectrum, and peak follower means which follows said spectrum and stores the peak value until the next subsequent scan of the spectrum.

7. A device according to claim 1 in which the background monitor has an energy envelope follower producing a voltage proportional to a sound energy input to the said device;

stable energy detector means which detects, for successive periods of the order of 40 milliseconds, whether the said voltage has remained inside given upper and lower threshold values; and

background level tracking means which compares an accumulated output from the stable energy detector means with the output of the energy envelope follower, so that when speech is present in the sound energy input, zero signal is provided by the background monitor to the computer.

8. A device according to claim 1 in which the unvoiced monitor includes a filter which removes frequencies lower than about 100 Hz;

an infinite clipper which clips the filter output;

a pulse width filter the output of which is arranged to be high at frequencies above about 2 KHz; and

logic and integrating means which integrates the output of the pulse width filter over periods of 20 milliseconds, whereby a signal is provided to the computer indicating whether the sound energy input has unvoiced characteristics.

9. A device according to claim 1 in which the voiced monitor includes a band pass filter passing frequencies in the range from about 40 Hz to about 2.5 KHz;

a tapped delay line fed from said filter, said delay line providing a maximum delay of the order of 15 milliseconds;

differential amplifier means for comparing the outputs from taps on the delay line with the input to the delay line, whereby all correspondences between the tap outputs and the input at any given time are matched to a resolution of the order of 156 microsecond;

circuit means to select continually the minimum and maximum values of said correspondences, and to compare the differences between said minimum and maximum values with two predetermined threshold values and hence provide to the computer a signal when the input sound has a fundamental repetitive structure; and

frequency measuring means which provides to the computer a measure of the periodicity of any said repetitive structure.

10. A method of supplying input information in a system of computer recognition of speech which comprises extracting information from an analogue signal which may contain speech and feeding in parallel to a computer, arranged for speech recognition, information derived from said analogue signal

(i) indicating whether the signal has a fundamental repetitive structure and the periodicity of such structure,

(ii) indicating whether the signal has no low frequency component and at the same time has high frequency components in a wide range of frequency,

(iii) indicating the presence and level of relatively stable energy in the signal, characteristic of background noise with an absence of speech,

(iv) indicating, in relation to a signal having no low frequency component, the peak frequency of the signal, the peak amplitude of the signal and the bandwidth, and

(v) indicating, in relation to a signal having a fundamental repetitive structure, the frequency of resonant content and the level of auto correlation thereof.