Nortel Networks Corporation

Northern Telecom Limited

Michael G Sabourin

Vishwa N Gupta

Serge Robillard

Marc A Fabiani

Real Tremblay

Jean-Francois Crespo

Waleed Fakhr

Kenneth W Smith

Jeff Marcus

The invention relates to a method and apparatus for training a multilingual speech model set. The multilingual speech model set generated is suitable for use by a speech recognition system for recognizing spoken utterances for at least two different languages. The invention allows using a single speech recognition unit with a single speech model set to perform speech recognition on utterances from two or more languages. The method and apparatus make use of a group of a group of acoustic sub-word units comprised of a first subgroup of acoustic sub-word units associated to a first language and a second subgroup of acoustic sub-word units associated to a second language where the first subgroup and the second subgroup share at least one common acoustic sub-word unit. The method and apparatus also make use of a plurality of letter to acoustic sub-word unit rules sets, each letter to acoustic sub-word unit rules set being associated to a different language. A set of untrained speech models is trained on the basis of a training set comprising speech tokens and their associated labels in combination with the group of acoustic sub-word units and the plurality of letter to acoustic sub-word unit rules sets. The invention also provides a computer readable storage medium comprising a program element for implementing the method for training a multilingual speech model set.

Method and apparatus for training a multilingual speech model set

The invention relates to a method and apparatus for generating transcriptions suitable for use in a speech-processing device. The invention provides processing the vocabulary item to derive a characteristic from the vocabulary item allowing to divide a pool of available languages in a first sub-group and a second sub-group. The vocabulary item manifests a higher probability of belonging to any one of the languages in the first sub-group than belonging to any language in the second sub-group. The invention further provides processing the vocabulary item to generate a group of transcriptions, the group of transcriptions being characterized by the absence of at least one transcription belonging to a language in the second sub-group of languages established for the vocabulary item. The group of transcriptions is then released for use by a speech-processing device.

Method and apparatus for generating multilingual transcription groups

The invention relates to a method and apparatus for generating phrasal transcriptions. The invention provides generating a group of word transcriptions for each vocabulary item in an orthographic phrase. According to a first embodiment, the invention further provides permuting the word transcriptions to generate a plurality of phrasal transcriptions and computing a score for each phrasal transcription in the plurality of phrasal transcriptions. The set of phrasal transcriptions is then selected from the plurality of phrasal transcriptions at least in part on a basis of the score data elements and stored in a format suitable for use by a speech recognition dictionary. As a variant, the phrasal transcriptions may be released in a format suitable for use by a speech synthesizer. According to a second embodiment, the invention further provides permuting the word transcriptions to generate the set of phrasal transcriptions, the permuting being characterized by yielding a higher likelihood of variability between the word transcriptions associated to a common vocabulary item among the set of phrasal transcriptions than a permuting rule using an outer-product procedure. The method further provides a computer readable medium comprising a program element operative to implement a method for generating a set of phrasal transcriptions.

Method and apparatus for generating phrasal transcriptions

A method and an apparatus for adding a new entry to a speech recognition dictionary, and more particularly a system and method for generating transcriptions from multiple utterances and a textual representation of a given word is provided. The method and apparatus combine acoustic information obtained from each of the training utterances to generate the most likely transcriptions. It also provides a method and apparatus for transcribing multiple utterances into a single transcription that can be added to a speech recognition dictionary. In a first step, a graph representative of possible pronunciation of the word or combination of words is generated on the basis of the textual representation of the word and a letter to phoneme set of rules. Phonemes perceptually close to the phonemes in the transcriptions obtained using the letter to phoneme rules are added on parallel branches in the generated graph. Following this, each branch in the graph is replaced with its allophones and the complete graph is scored using training utterances to obtain the most likely allophone sequence on the basis of the multiple utterances of the word.

Method and apparatus for obtaining a transcription of phrases through text and spoken utterances

An adaptive speech recognition system is provided including an input for receiving a signal derived from a spoken utterance indicative of a certain vocabulary item, a speech recognition dictionary, a speech recognition unit and an adaptation module. The speech recognition dictionary has a plurality of vocabulary items each being associated to a respective dictionary transcription group. The speech recognition unit is in an operative relationship with the speech recognition dictionary and selects a certain vocabulary item from the speech recognition dictionary as being a likely match to the signal received at the input. The results of the speech recognition process are provided to the adaptation module. The adaptation module includes a transcriptions bank having a plurality of orthographic groups, each including a plurality of transcriptions associated with a common vocabulary item. A transcription selector module in the adaptation module retrieves a given orthographic group from the transcriptions bank on a basis of the vocabulary item recognized by the speech recognition unit. The transcription selector module processes the given orthographic group on the basis of the signal received at the input to select a certain transcription from the transcriptions bank. The adaptation module then modifies a dictionary transcription group corresponding to the vocabulary item selected as being a likely match to the signal received at the input on the basis of the selected certain transcription.

Method and apparatus for providing unsupervised adaptation of transcriptions

An automatic transcription tool executing on a computer system includes a text pre-processor, a phrasal parser, a word transcriber, and a phrasal transcription post-processor. Word orthographies are input to the text pre-processor and are then sequentially processed by the phrasal parser, word transcriber, and phrasal transcription post-processor. In general, the text pre-processor performs initial substitution processing on the input orthography sequences, the phrasal parser groups related adjacent input orthographies into clauses, the word transcriber performs phonetic transcription, and the phrasal transcription post-processor performs functions, such as, liasion handling and sanity checking on the output phonetic transcriptions. The phonemic transcription of the input orthography is eventually output.

Automatic transcription tool

A confusability tool generates a confusability cost associated with two phonemic transcriptions. The confusability cost measures the likelihood that a human or machine hearing the first word will mistakenly hear the second word. The cost calculation is based on a weighting of the Levinstein distance between the transcription pair.

Predicting auditory confusions using a weighted Levinstein distance

Systems and methods consistent with the present invention determine whether to accept one of a plurality of intermediate recognition results output by a speech recognition system as a final recognition result. The system first combines a plurality of speech rejection features into a feature function in which weights are assigned to each rejection feature in accordance with a recognition accuracy of each rejection feature. Feature values are then calculated for each of the rejection features using the plurality of intermediate recognition results. The system next computes the feature function according to the calculated feature values to determine a rejection decision value. Finally, one of the plurality of intermediate recognition results is accepted as the final recognition result according to the rejection decision value.

Speech recognition rejection method using generalized additive models

A vocabulary sub-set is selected from a large speech recognition dictionary. The selected vocabulary sub-set may be used in a real time directory assistance system to improve the system's real-time performance. The selection process is effected on the basis of the cost-benefit ratio, the benefit being measured in savings in operator working time. On the other hand, the cost is measured in terms of hardware limitations, namely processor throughput. Typically, the vocabulary sub-set is limited to a maximum number orthographies that would enable the system to achieve real-time performance.

Method and apparatus for selecting a vocabulary sub-set from a speech

The invention relates to a method and apparatus for automatically generating a speech recognition vocabulary for a speech recognition system from a listing that contains a number of entries, each entry containing a multi-word identification data that distinguishes that entry from other entries in the list. The method comprises the steps of creating for each entry in the listing a plurality of orthographies in the speech recognition vocabulary that are formed by combining selected words from the entry. The words combination is effected by applying a heuristics model that mimics the way users formulate requests to the automated directory assistance system. The method is particularly useful for generating speech recognition vocabularies for automated directory assistance systems.

Growing community of inventors

St. Lambert, Canada

Michael G Sabourin