David K. Burton

Discrete utterance speech recognition without time alignment

The results of a new method are presented for discrete utterance speech recognition. The method is based on rate-distortion speech coding (speech coding by vector quantization), minimum cross-entropy pattern classification, and information-theoretic spectral distortion measures. Separate vector quantization code books are designed from training sequences for each word in the recognition vocabulary. Inputs from outside the training sequence are classified by performing vector quantization and finding the code book that achieves the lowest average distortion per speech frame. The new method obviates time alignment. It achieves 99 percent accuracy for speaker-dependent recognition of a 20 -word vocabulary that includes the ten digits, with higher accuracy for recognition of the digit subset. For speaker-independent recognition, the method achieves 88 percent accuracy for the 20 -word vocabulary and 95 percent for the digit subset. Background of the method, detailed empirical results, and an analysis of computational requirements are presented.

Isolated-word speech recognition using multisection vector quantization codebooks

A new approach to isolated-word speech recognition using vector quantization (VQ) is examined. In this approach, words are recognized by means of sequences of VQ codebooks, called multisection codebooks. A separate multisection codebook is designed for each word in the recognition vocabulary by dividing the word into equal-length sections and designing a standard VQ codebook for each section. Unknown words are classified by dividing them into corresponding sections, encoding them with the multisection codebooks, and finding the multisection codebook that yields the smallest average distortion. For speaker-independent recognition of the digits, this approach achieved a recognition accuracy of 98 percent. In addition, the approach achieved greater than 99 percent accuracy for speaker-dependent recognition of the digits with only one distortion computation per input frame per vocabulary word. The approach is described, detailed experimental results are presented and discussed, and computational requirements are analyzed.

Text-dependent speaker verification using vector quantization source coding

Several vector quantization approaches to the problem of text-dependent speaker verification are described. In each of these approaches, a source codebook is designed to represent a particular speaker saying a particular utterance. Later, this same utterance is spoken by a speaker to be verified and is encoded in the source codebook representing the speaker whose identity was claimed. The speaker is accepted if the verification utterances quantization distortion is less than a prespecified speaker-specific threshold. The best approach achieved a 0.7 percent false acceptance rate and a 0.6 percent false rejection rate on a speaker population comprising 16 admissible speakers and 111 casual imposters. The approaches are described, and detailed experimental results are presented and discussed.

Text-dependent speaker recognition using vector quantization

An application of source coding to speaker recognition is described. The method is text-dependent - the text spoken is known, and the problem is to determine who said it. Each speaker is represented by a sequence of vector quantization codebooks; known input utterances are classified using these codebook sequences and the resulting classification distortion is compared to a rejection threshold. On a 16 speaker test population with an additional 111 imposters, this method achieved a false rejection rate of 0.8%, an imposter acceptance rate of 1.8%, and within the 16 speakers, an identification error rate of 0.0%.

Discrete utterance speech recognition without time normalization

We present a new, fast method for discrete utterance recognition of telephone bandwidth speech. The method is based on speech coding by vector quantization and minimum cross-entropy pattern classification. Separate vector quantization codebooks are designed from training sequences for each word in the recognition vocabulary. Inputs from outside the training sequence are classified by performing vector quantization and finding the codebook that achieves the lowest average distortion per speech frame. The new method obviates time normalization and uses approximately 6000 bits to represent each utterance in the recognition vocabulary. Preliminary limited testing on speaker dependent digit recognition has demonstrated excellent performance. Detailed tests are now in progress.

Applying matrix quantization to isolated word recognition

A new approach to isolated word recognition is examined. This approach is based on an extension of vector quantization speech coding, called matrix quantization speech coding, that was developed by Tsao and Gray. In this new approach, a codebook containing a set of time-ordered-sequences of speech spectra represents each vocabulary word. A word is recognized by encoding it with each codebook and classifying the input word according to the codebook that yields the smallest distortion. On the digits, this approach achieved a speaker independent recognition accuracy greater than 98%. The approach is described, experimental results are presented, and comparisons with vector quantization based approaches are given.

Parameter selection for isolated word recognition using vector quantization

The use of vector quantization (VQ) in isolated-word recognition of a 20-word vocabulary is examined. A separate sequence of VQ code books is designed for each word in the recognition vocabulary and input words are classified by performing VQ and finding the sequence of code books that achieve the smallest average distortion. In this paper, critical parameters are noted and the results of parameter studies are presented.

Speaker-dependent isolated word recognition using speaker-independent vector quantization codebooks augmented with speaker-specific data

A hybrid approach to speaker-dependent isolated word recognition is discussed. The approach merges speaker-specific information obtained from a single training utterance with multisection vector quantization codebooks that were designed for speaker-independent recognition. The approach provides easily trained, computationally efficient, and accurate isolated word recognition. On the digits, the approach achieved an error rate less than 1 percent.

Speech noise reduction by means of multi-signal minimum-cross-entropy spectral analysis

This paper presents results of a new spectrum-analysis method that estimates a number of power spectra when a prior estimate of each is available and new information is obtained in the form of values of the auto-correlation function of their sum. The method applies for instance when one obtains autocorrelation measurements for a signal with independent additive interference, and one has prior estimates of the signal and noise spectra. By incorporating prior estimates for both spectra, the method offers considerable flexibility for tailoring an estimator to the characteristics of a signal or noise. The new method, a generalization of Minimum-Cross-Entropy Spectrum Analysis (MCESA) and Maximum Entropy Spectrum Analysis (MESA), is called Multisignal MCESA. Its theoretical basis is reviewed, and results of experimental tests of an implementation are presented. The test data comprise digitized samples of speech corrupted with helicopter noise and tone interference.

Acoustic transient classification of passive sonar signals by using vector quantization

The automatic classifications of ice-generated transient signals is examined. The classification method is a modified version of one previously used to classify speech sounds and is based on vector quantization. It uses a clustering algorithm on training data to design a signal-specific classification codebook (a set of reference spectra) for each signal type, and it compares an unknown signal event with each of the classification codebooks to make a classification decision. On a test set consisting 7 types of ice-generated signals, a recognition accuracy of 71% was achieved. Details of the experimental conditions and results are given, and recommendations for improving the performance are provided.<<ETX>>