International Business Machines Corporation

Oipenn, Inc.

Miroslav Novak

Michael Alan Picheny

George Andrei Saon

Mark Edward Epstein

Raimo Bakis

Kenneth W Church

Dimitrios B Dimitriadis

William Stuart Meisel

Petr Fousek

Ridley M Whitaker

Martin Franz

An approach is provided that receives an audio stream and utilizes a voice activation detection (VAD) process to create a digital audio stream of voices from at least two different speakers. An automatic speech recognition (ASR) process is applied to the digital stream with the ASR process resulting in the spoken words to which a speaker turn detection (STD) process is applied to identify a number of speaker segments with each speaker segment ending at a word boundary. A speaker clustering algorithm is then applied to the speaker segments to associate one of the speakers with each of the speaker segments.

Diarization driven by the ASR based segmentation

An approach is provided that receives an audio stream and utilizes a voice activation detection (VAD) process to create a digital audio stream of voices from at least two different speakers. An automatic speech recognition (ASR) process is applied to the digital stream with the ASR process resulting in the spoken words to which a speaker turn detection (STD) process is applied to identify a number of speaker segments with each speaker segment ending at a word boundary. The STD process analyzes a number of speaker segments using a language model that determines when speaker changes occur. A speaker clustering algorithm is then applied to the speaker segments to associate one of the speakers with each of the speaker segments.

Diarization driven by meta-information identified in discussion content

A multi-lingual translation system that provides multiple output sentences for a given word or phrase. Each output sentence for a given word or phrase reflects, for example, a different emotional emphasis, dialect, accents, loudness or rates of speech. A given output sentence could be selected automatically, or manually as desired, to create a desired effect. For example, the same output sentence for a given word or phrase can be recorded three times, to selectively reflect excitement, sadness or fear. The multi-lingual translation system includes a phrase-spotting mechanism, a translation mechanism, a speech output mechanism and optionally, a language understanding mechanism or an event measuring mechanism or both. The phrase-spotting mechanism identifies a spoken phrase from a restricted domain of phrases. The language understanding mechanism, if present, maps the identified phrase onto a small set of formal phrases. The translation mechanism maps the formal phrase onto a well-formed phrase in one or more target languages. The speech output mechanism produces high-quality output speech. The speech output may be time synchronized to the spoken phrase using the output of the event measuring mechanism.

Method and apparatus for translating natural-language speech using multiple output phrases

A multi-lingual time-synchronized translation system and method provide automatic time-synchronized spoken translations of spoken phrases. The multi-lingual time-synchronized translation system includes a phrase-spotting mechanism, optionally, a language understanding mechanism, a translation mechanism, a speech output mechanism and an event measuring mechanism. The phrase-spotting mechanism identifies a spoken phrase from a restricted domain of phrases. The language understanding mechanism, if present, maps the identified phrase onto a small set of formal phrases. The translation mechanism maps the formal phrase onto a well-formed phrase in one or more target languages. The speech output mechanism produces high-quality output speech using the output of the event measuring mechanism for time synchronization. The event-measuring mechanism measures the duration of various key events in the source phrase. Event duration could be, for example, the overall duration of the input phrase, the duration of the phrase with interword silences omitted, or some other relevant durational features. The present invention recognizes the quality improvements can be achieved by restricting the task domain under consideration.

Method and apparatus for time-synchronized translation and synthesis of natural-language speech

A method for fast match processing, comprising two stages, a pre-processing stage and an on-line stage. The pre-processing stage comprises the steps of computing an a-priori probability of occurrence for each word from an acoustic vocabulary; deriving a penalty score for each word from said acoustic vocabulary based on each words a-priori probability of occurrence in an input text. The on-line stage operates on an input text stream, comprising the steps of, computing a path score for each word from said input text; combining the computed path score with the derived penalty score to form a combined score and testing the combined score against a threshold to determine top ranking candidate words.

Non-leaf node penalty score assignment system and method for improving acoustic fast match speed in large vocabulary systems

A method is provided for reducing search complexity in a speech recognition system having a fast match, a detailed match, and a language model. Based on at least one predetermined variable, the fast match is optionally employed to generate candidate words and acoustic scores corresponding to the candidate words. The language model is employed to generate language model scores. The acoustic scores are combined with the language model scores and the combined scores are ranked to determine top ranking candidate words to be later processed by the detailed match, when the fast match is employed. The detailed match is employed to generate detailed match scores for the top ranking candidate words.

Method for reducing search complexity in a speech recognition system

Methods and apparatus for performing a tree search based acoustic fast match in a speech recognition system for decoding a speech utterance, the tree having a tree root and tree nodes connected by tree branches, the tree nodes having phonetic models associated therewith, are provided. An illustrative embodiment of the method comprises: providing a cache having cache cells for storing phone probabilities therein; selecting a first branch leading to a next node, said branch selection starting at the tree root; accessing the cache to select a particular cache cell where the probability of a particular match is stored; evaluating the phonetic model to obtain the probability and storing the probability and an associated end time in the cache cell, if the cache cell accessed in the accessing step does not contain the required probability; using the probability value and the associated end time stored in the cache cell, if the cache cell accessed in the accessing step contains the required probability; selecting a new branch to proceed to the next node; and iteratively continuing from the accessing step until the whole tree is traversed and all possible word candidates associated with the speech recognition system are evaluated.

Growing community of inventors

Mohegan Lake, NY, United States of America

Miroslav Novak