Efficient empirical determination, computation, and use of acoustic confusability measures

• US 10,121,469 B2
• Filed: 03/13/2017
• Issued: 11/06/2018
• Est. Priority Date: 10/31/2002
• Status: Active Grant

• Current Assignee: Promptu Systems Corporation
• Original Assignee: Promptu Systems Corporation
• Inventors: Printz, Harry, Chittar, Naren
• Primary Examiner(s): Lerner, Martin
• Application Number:
  US15/457,964
• Publication Number:
  US 20170186421A1
• Time in Patent Office: 603 Days
• Field of Search: 704235, 704236, 704242, 704243, 7042562
• CPC Class Codes: G06F 16/95 Retrieval from the web
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 17/26 Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

Efficient empirical determination, computation, and use of an acoustic confusability measure comprises: (1) an empirically derived acoustic confusability measure, comprising a means for determining the acoustic confusability between any two textual phrases in a given language, where the measure of acoustic confusability is empirically derived from examples of the application of a specific speech recognition technology, where the procedure does not require access to the internal computational models of the speech recognition technology, and does not depend upon any particular internal structure or modeling technique, and where the procedure is based upon iterative improvement from an initial estimate; (2) techniques for efficient computation of empirically derived acoustic confusability measure, comprising means for efficient application of an acoustic confusability score, allowing practical application to very large-scale problems; and (3) a method for using acoustic confusability measures to make principled choices about which specific phrases to make recognizable by a speech recognition application.

Claims

1. A method for generating an acoustic confusability measure, said method comprising the steps of:

receiving as input a corpus, comprising a set of utterances with corresponding reliable transcriptions;

recognizing, via an automatic speech recognition system, at least one utterance among the set of utterances to yield a recognized utterance, wherein said recognized utterance includes at least one decoded frame sequence;

coalescing identical sequential phonemes of said at least one decoded frame sequence to yield at least one decoded phoneme sequence;

determining, for each corresponding reliable transcription, at least one pronunciation, wherein said at least one pronunciation includes at least one true phoneme sequence;

generating as output said recognized corpus comprising for each said recognized utterance at least said at least one decoded phoneme sequence and said at least one true phoneme sequence; and

generating an empirically derived acoustic confusability measure from said recognized corpus, said empirically derived acoustic confusability measure comprising a family of probability models Π

={p(d|t)} wherein each of d and t are phonemes drawn from an augmented phoneme alphabet Φ

′

.

2. The method of claim 1, wherein said recognizing, via an automatic speech recognition system, at least one utterance is performed in an N-best manner, wherein up to N variant decodings are produced for each recognized utterance, and wherein each said variant decoding includes a distinct decoded frame sequence.

3. The method of claim 2, wherein each said variant decoding includes a corresponding confidence score.

4. The method of claim 3, wherein in said method for generating said empirically derived acoustic confusability measure, a count associated with processing each said variant decoding is incremented by said corresponding confidence score.

5. The method of claim 1, further comprising:

applying a phoneme map to frames of said at least one decoded frame sequence to yield at least one mapped decoded frame sequence, said at least one mapped decoded frame sequence thereafter replacing said at least one decoded frame sequence.

6. The method of claim 5, in which said applying the phoneme map comprises at least one of the steps of:

mapping a phoneme of each frame of said at least one decoded frame sequence to yield a mapped phoneme, said mapped phoneme being either the same as or different from an original phoneme, thereby yielding said at least one decoded frame sequence;

or

mapping said phoneme of each frame of said at least one decoded frame sequence to yield said mapped phoneme, said mapped phoneme being either the same as or different from the original phoneme, to include left phonetic context, right phonetic context or both into the mapped phoneme, thereby yielding said at least one decoded frame sequence.

7. The method of claim 1, wherein said determining for each reliable transcription said at least one pronunciation comprises any of the steps of:

for each word w of said each reliable transcription, utilizing as a pronunciation of w, from a set of all pronunciations of w, its most popular pronunciation;

for each word w of said each reliable transcription, utilizing as the pronunciation of w, from the set of all pronunciations of w, a pronunciation selected at random;

for each word w of said each reliable transcription, utilizing as the pronunciation of w, from the set of all pronunciations of w, the pronunciation that is closest, in the sense of string edit distance, to the at least one decoded phoneme sequence for w within the said at least one decoded phoneme sequence;

or

for each word w of said each reliable transcription, utilizing as the pronunciation of w each of a plurality of pronunciations from the set of all pronunciations of w, thereby generating a multiplicity of pronunciations for each reliable transcription.