Speech recognition utilizing multitude of speech features

• US 20050119885A1
• Filed: 11/28/2003
• Published: 06/02/2005
• Est. Priority Date: 11/28/2003
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Kuo, Hong-Kwang, Balakrishnan, Sreeram Viswanath, Nahamoo, David, Gopinath, Ramesh A., Picheny, Michael Alan, Zweig, Geoffrey G., Maison, Benoit, Chen, Stanley F., Saon, George A., Axelrod, Scott E., Gao, Yuging
• Granted Patent: US 7,464,031 B2
• Time in Patent Office: Days
• US Class Current: 704/231
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/14 using statistical models, e.g. Hidden Markov Models [HMMs] G10L15/18 takes precedence
• CPC Class Codes: G10L 2015/085 Methods for reducing search complexity, pruning

Abstract

In a speech recognition system, the combination of a log-linear model with a multitude of speech features is provided to recognize unknown speech utterances. The speech recognition system models the posterior probability of linguistic units relevant to speech recognition using a log-linear model. The posterior model captures the probability of the linguistic unit given the observed speech features and the parameters of the posterior model. The posterior model may be determined using the probability of the word sequence hypotheses given a multitude of speech features. Log-linear models are used with features derived from sparse or incomplete data. The speech features that are utilized may include asynchronous, overlapping, and statistically non-independent speech features. Not all features used in training need to appear in testing/recognition.

Claims

1. A speech recognition system, comprising:

a features extractor that extracts a multitude of speech features;

a log-linear function that receives the multitude of speech features to determine a posterior probability of a hypothesized linguistic unit given the extracted multitude of speech features, and a search device that consults the log-linear function to determine a recognized output of unknown utterances.

2. The speech recognition system of claim 1, wherein the log linear function models the posterior probability using a log linear model

3. The speech recognition system of claim 1, wherein the speech features comprise at least one of asynchronous, overlapping, and statistically non-independent speech features.

4. The speech recognition system of claim 1, wherein at least one of the speech features extracted is derived from incomplete data.

5. The speech recognition system of claim 1, further comprising a loopback.

6. The speech recognition system of claim 1, wherein the features are extracted using direct matching between test data and training data.

7. A speech recognition method, comprising:

extracting a multitude of speech features;

determining a posterior probability of a hypothesized linguistic unit given the extracted multitude of speech features, and using a log-linear function, determining a recognized output of unknown utterances.

8. The speech recognition method of claim 7, wherein the log linear function models the posterior probability using a log linear model.

9. The speech recognition method of claim 7, wherein the speech features comprise at least one of asynchronous, overlapping, and statistically non-independent speech features.

10. The speech recognition method of claim 7, wherein at least one of the speech features extracted is derived from incomplete data.

11. The speech recognition method of claim 7, further comprising a step of loopback.

12. The speech recognition method of claim 7, wherein the features are extracted using direct matching between test data and training data.