Quantization using frequency and mean compensated frequency input data for robust speech recognition

• US 6,219,642 B1
• Filed: 10/05/1998
• Issued: 04/17/2001
• Est. Priority Date: 10/05/1998
• Status: Expired due to Term

• Current Assignee: RPX Corporation
• Original Assignee: Legerity Incorporated (Microsemi Semiconductor Corp.)
• Inventors: Cong, Lin, Asghar, Safdar M.
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): ARMSTRONG, ANGELA A
• Application Number:
  US09/166,648
• Time in Patent Office: 925 Days
• Field of Search: 704/230, 704/231, 704/226, 704/256, 704/206, 704/209, 704/243
• US Class Current: 704/256.8
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/144 Training of HMMs
• 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

A speech recognition system utilizes multiple quantizers to process frequency parameters and mean compensated frequency parameters derived from an input signal. The quantizers may be matrix and vector quantizer pairs, and such quantizer pairs may also function as front ends to a second stage speech classifiers such as hidden Markov models (HMMs) and/or utilizes neural network postprocessing to, for example, improve speech recognition performance. Mean compensating the frequency parameters can remove noise frequency components that remain approximately constant during the duration of the input signal. HMM initial state and state transition probabilities derived from common quantizer types and the same input signal may be consolidated to improve recognition system performance and efficiency. Matrix quantization exploits the “evolution” of the speech short-term spectral envelopes as well as frequency domain information, and vector quantization (VQ) primarily operates on frequency domain information. Time domain information may be substantially limited which may introduce error into the matrix quantization, and the VQ may provide error compensation. The matrix and vector quantizers may split spectral subbands to target selected frequencies for enhanced processing and may use fuzzy associations to develop fuzzy observation sequence data. A mixer may provide a variety of input data to the neural network for classification determination. Fuzzy operators may be utilized to reduce quantization error. Multiple codebooks may also be combined to form single respective codebooks for split matrix and split vector quantization to reduce processing resources demand.

Claims

1. An apparatus comprising:

a mean compensation module to receive parameters for TO samples of an input signal and to generate mean compensated parameters from the received input signal parameters;

a first quantizer to receive the input signal parameters and to quantize the input signal parameters, wherein the first quantizer is comprised of a first matrix quantizer and a first vector quantizer;

a second quantizer to receive the input signal mean compensated parameters and to quantize the input signal mean compensated parameters wherein the second quantizer is comprised of a second matrix quantizer and a second vector quantizer;

a backend processor to receive the quantized input signal parameters and the input signal mean compensated input signal parameters and to classify the input signal therefrom, wherein the backend processor comprises;

a first group of hidden Markov models which are trained using quantized input signal parameters from the first and second matrix quantizers;

a second group of hidden Markov models which are trained using quantized input signal parameters from the first and second vector quantizers; and

a stochastic module to (a) receive the quantized input signal parameters from the first and second matrix quantizers, (b) determine the respective probabilities that each hidden Markov model from the first group of hidden Markov models modeled the quantized input signal parameters from the first and second matrix quantizers, (c) receive the quantized input signal parameters from the first and second vector quantizers, (d) determine the respective probabilities that each hidden Markov model from the second group of hidden Markov models to have modeled the quantized input signal parameters from the first and second vector quantizers, wherein the backend processor is capable of utilizing the probabilities to generate the input signal classification.

2. The apparatus as in claim 1 wherein the backend processor further comprises:

decision logic to receive the respective probabilities and to combine the respective probabilities to classify the input signal therefrom.

3. The apparatus as in claim 1 wherein the backend processor further comprises:

a mixer to combine the respective probabilities;

a neural network to receive the combined respective probabilities and to determine respective likelihood data which relate the input data to a vocabulary word of the apparatus; and

decision logic to receive the likelihood data and to classify the input signal therefrom.

4. The apparatus as in claim 1 wherein the input signal parameters are frequency parameters.

5. The apparatus as in claim 1 wherein the vector quantizer in the first quantizer is capable of generating vector quantized data and the matrix quantizer in the first quantizer is capable of generating matrix quantized data, wherein the vector quantized data is capable of being combined with the matrix quantized data to generate the quantized input signal frequency parameters.

6. The apparatus as in claim 1 wherein the vector quantizer in the second quantizer is capable of generating mean compensated vector quantized data and the matrix quantizer in the second quantizer is capable of generating mean compensated matrix quantized data, wherein the mean compensated vector quantized data is capable of being combined with the mean compensated matrix quantized data to generate the quantized input signal mean compensated frequency parameters.

7. The apparatus as in claim 1 wherein the stochastic module further comprises a Viterbi algorithm.