Lucent Technologies Inc.

At+t Corp.

Agere Systems Guardian Corp.

Wu Chou

Biing-Hwang Juang

Chin-Hui Lee

Mazin G Rahim

Eric Rolse Buhrke

Nambirajan Seshadri

Tatsuo Matsuoka

Michael Charles Recchione

Qiru Zhou

Eric Rolfe Buhrke

Tatsuya Kawahara

A continuous speech decoder that is built up of multiple layers. Each of the layers uses independent knowledge sources and rules, but all the layers cooperate to quickly decode the speech input into words. A first layer is concerned with acoustic data, a second layer with phone data of speech and a third layer concerns word data and word sequences. By separating these layers, the higher layers can be made time independent and asynchronous. Thus the asynchronous layers can process data quickly and give fast support to the first layer which keeps a dynamic record called a dynamic network of the most likely continuous speech results. The speed and separation of this decoder allows better memory efficiency and better decoder results compared to previously known continuous speech decoders.

Method and apparatus for continuous speech recognition using a layered, self-adjusting decoded network

Automatic speech/speaker recognition over digital wireless channels

Method and apparatus for continuous speech recognition using a layered,

Hierarchical signal bias removal (HSBR) signal conditioning uses a codebook constructed from the set of recognition models and is updated as the recognition models are modified during recognition model training. As a result, HSBR signal conditioning and recognition model training are based on the same set of recognition model parameters, which provides significant reduction in recognition error rate for the speech recognition system.

Signal conditioned minimum error rate training for continuous speech

Disclosed are systems, methods and articles of manufacture for performing high resolution N-best string hypothesization during speech recognition. A received input signal, representing a speech utterance, is processed utilizing a plurality of recognition models to generate one or more string hypotheses of the received input signal. The plurality of recognition models preferably include one or more inter-word context dependent models and one or more language models. A forward partial path map is produced according to the allophonic specifications of at least one of the inter-word context dependent models and the language models. The forward partial path map is traversed in the backward direction as a function of the allophonic specifications to generate the one or more string hypotheses. One or more of the recognition models may represent one phone words.

Systems, methods and articles of manufacture for performing high

A key-phrase detection and verification method that can be advantageously used to realize understanding of flexible (i.e., unconstrained) speech. A 'multiple pass' procedure is applied to a spoken utterance comprising a sequence of words (i.e., a 'sentence'). First, a plurality of key-phrases are detected (i.e., recognized) based on a set of phrase sub-grammars which may, for example, be specific to the state of the dialogue. These key-phrases are then verified by assigning confidence measures thereto and comparing these confidence measures to a threshold, resulting in a set of verified key-phrase candidates. Next, the verified key-phrase candidates are connected into sentence hypotheses based upon the confidence measures and predetermined (e.g., task-specific) semantic information. And, finally, one or more of these sentence hypotheses are verified to produce a verified sentence hypothesis and, from that, a resultant understanding of the spoken utterance.

Method of key-phase detection and verification for flexible speech

A joint data (features) and channel (bias) estimation framework for robust processing of speech received over a channel is described. A trellis encoded vector quantizer is used as a pre-processor to estimate the channel bias using blind maximum likelihood sequence estimation. Sequential constraint in the feature vector sequence of a speech signal is applied for the selection of the quantized signal constellation and for the decoding process in joint data and channel estimation. A two state trellis encoded vector quantizer is designed for signal bias removal applications.

Speech coding and joint data/channel bias estimation using finite state

In a speech recognition system, a recognition processor receives an unknown utterance signal as input. The recognition processor in response to the unknown utterance signal input accesses a recognition database and scores the utterance signal against recognition models in the recognition database to classify the unknown utterance and to generate a hypothesis speech signal. A verification processor receives the hypothesis speech signal as input to be verified. The verification processor accesses a verification database to test the hypothesis speech signal against verification models reflecting a preselected type of training stored in the verification database. Based on the verification test, the verification processor generates a confidence measure signal. The confidence measure signal can be compared against a verification threshold to determine the accuracy of the recognition decision made by the recognition processor.

Discriminative utterance verification for connected digits recognition

A method of making a speech recognition model database is disclosed. The database is formed based on a training string utterance signal and a plurality of sets of current speech recognition models. The sets of current speech recognition models may include acoustic models, language models, and other knowledge sources. In accordance with an illustrative embodiment of the invention, a set of confusable string models is generated, each confusable string model comprising speech recognition models from two or more sets of speech recognition models (such as acoustic and language models). A first scoring signal is generated based on the training string utterance signal and a string model for that utterance, wherein the string model for the utterance comprises speech recognition models from two or more sets of speech recognition models. One or more second scoring signals are also generated, wherein a second scoring signal is based on the training string utterance signal and a confusable string model. A misrecognition signal is generated based on the first scoring signal and the one or more second scoring signals. Current speech recognition models are modified, based on the misrecognition signal to increase the probability that a correct string model will have a rank order higher than other confusable string models.

Minimum error rate training of combined string models

A system pattern-based speech recognition, e.g., a hidden Markov model (HMM) based speech recognizer using Viterbi scoring. The principle of minimum recognition error rate is applied by the present invention using discriminative training. Various issues related to the special structure of HMMs are presented. Parameter update expressions for HMMs are provided.

Growing community of inventors

Berkeley Heights, NJ, United States of America

Wu Chou