Apparatus and methods for the selective addition of noise to templates employed in automatic speech recognition systems

• US 4,933,973 A
• Filed: 08/16/1989
• Issued: 06/12/1990
• Est. Priority Date: 02/29/1988
• Status: Expired due to Term

• Current Assignee: ITT Corporation (ITT, Inc.)
• Original Assignee: ITT Corporation (ITT, Inc.)
• Inventors: Porter, Jack E.
• Primary Examiner(s): Kemeny, Emanuel S.
• Application Number:
  US07/395,211
• Time in Patent Office: 300 Days
• Field of Search: 381/43, 381/46, 381/47
• 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
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• CPC Class Codes: G10L 21/0216 characterised by the method used for estimating noise
• CPC Class Codes: G10L 25/18 the extracted parameters being spectral information of each sub-band

Abstract

To improve the recognition of incoming speech signals in noise, the prestored templates of noise-free speech are modified to have the estimated spectral values of noise and the same signal-to-noise ratio as the incoming signal.

Claims

1. In a speech recognition system of the type including a storage for storing initial templates representing spectral values of recognizable speech in the absence of noise, a spectrum analyzer for providing spectral values of utterances of an incoming signal representing speech in the presence of noise at an output thereof, and a recognition module for comparing operational templates with the output spectral values from said spectrum analyzer to provide an output upon a favorable comparison indicative of the presence of recognized speech in said utterances, the improvement therewith of apparatus for generating said operation templates, comprising:

first means coupled to said spectrum analyzer for providing an estimated noise signal indicative of the noise in the incoming signal, and second means coupled to said first means and responsive to said estimated noise signal to generate said operational templates which are modified from said initial templates according to said estimated noise signal,

wherein said first means includes a speech and noise level tracking means having a speech tracking portion operative to detect a speech signal in the presence of noise in said utterances and to provide at one output thereof a first signal indicative of an average power of the speech signal in the presence of noise which is a scalar speech level value associated with the speech signal, and a noise tracking portion operative to detect noise in said utterances and to provide a given time period of said utterances which is a vector of spectral values representing an estimate of the noise, and

wherein said second means generates said operational templates by adjusting the speech level of said initial templates in accordance with said first signal and by adding spectral values of the estimate of the noise in accordance with said second signal to said initial templates,

whereby the operational templates have the estimated noise of and the same signal-to-noise ratio as the utterances of the incoming signal in order to obtain an improved speech recognition performance of said recognition module.

2. The speech recognition system according to claim 1, wherein said spectrum analyzer comprises a plurality of bandpass filters arranged in a filter bank array with each filter adapted to pass a given spectral component according to the bandwidth of the said filter.

3. The speech recognition system according to claim 2, wherein said first means includes means for measuring the average and the variance of said bandpass filters to provide an estimate of the noise passing properties of each filter.

4. The speech recognition system according to claim 3, wherein said noise estimate is estimated on the basis of said filter response to Gaussian noise.

5. The speech recognition system according to claim 1, wherein said templates generated in the absence of noise are noise free token templates and means responsive to said templates to provide an average value to provide at outputs Base Form data, and means for modifying said Base Form data according to a current predicted noise signal.

6. A speech recognition system according to claim 1, further comprising:

processing means coupled to said spectrum analyzer for generating said operational templates for storage by modifying said initial templates according to said estimated noise signal indicative of the presence of noise.

7. The speech recognition system according to claim 6, wherein said processing means said expected calculated value is indicative of the presence of Gaussian noise.

8. The speech recognition system according to claim 6, wherein said processing means includes means for averaging noise-free templates to provide Base Form data outputs and modifying said Base Form data outputs by adding to said data, noise data calculated.

9. The speech recognition system according to claim 6, wherein said processing means includes averaging means for providing at output the average value of successive pairs of said spectral magnitude values as provided by said analyzer,

scaling means coupled to said averaging means output for providing a given length field signal and means for converting said given length field signal to a logarithmic signal for providing one of said Base Form data outputs.

10. The speech recognition system according to claim 9, further including,

squaring means coupled to said averaging means for providing at an output a vector signal indicative of the squared magnitude of said average value of successive pairs and means coupled to the output of said squaring means for providing other ones of said Base Form data outputs.

11. The speech recognition system according to claim 10, wherein said means coupled to the output of said squaring means includes relative energy forming means responsive to said vector signal to provide a Base Form energy parameter and speech and noise level tracker means for providing at an output a Base Form parameter indicative of the power level of both speech and noise.

12. A method of forming operational templates for use in a speech recognition system for recognizing speech in the presence of noise in utterances of an incoming signal based upon comparison of spectral values thereof to the operational templates, comprising the steps of:

providing an estimated noise signal by detecting a speech signal in the presence of noise in the incoming signal and providing a first signal indicative of an average power of the speech signal in the presence of noise which is a scalar speech level value associated with the speech signal, and detecting noise in said utterances and providing a second signal indicative of an average power of the noise over a given time period of said utterances which is a vector of spectral values representing an estimate of the noise, and

modifying initial templates representing recognizable speech in the absence of noise by adjusting the speech level of said initial templates in accordance with said first signal and by adding spectral values of the estimate of the noise in accordance with said second signal to said initial templates, in order to form said operational templates having the estimated noise of and the same signal-to-noise ratio as the utterances of the incoming signal for obtaining an improved speech recognition performance.

13. The method according to claim 12, wherein said step of providing includes measuring the response of a given speech processing channel with respect to noise and estimating said signal to be provided based on said measurement.

14. The method according to claim 12, wherein said step of modifying includes first forming a Base Form template relatively free from noise and,

modifying said Base Form template according to said signal indicative of said expected noise level.

15. The method according to claim 12, wherein the step of modifying includes forming Base Form templates relatively free from noise,

adding noise to each template and,

averaging said added noise template data to form new templates according to said analyzed data.