Method for constructing a model of a new word for addition to a word model database of a speech recognition system

• US 5,873,061 A
• Filed: 05/02/1996
• Issued: 02/16/1999
• Est. Priority Date: 05/03/1995
• Status: Expired due to Fees

• Current Assignee: US Philips Corporation (Koninklijke Philips N.V.)
• Original Assignee: US Philips Corporation (Koninklijke Philips N.V.)
• Inventors: Thelen, Eric, Beyerlein, Peter, Hab-Umbach, Reinhold
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US08/642,018
• Time in Patent Office: 1,020 Days
• Field of Search: 395/2.6, 395/2.63, 395/2.64, 395/2.65, 395/2.4
• US Class Current: 704/254
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/06 Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice G10L15/14 takes precedence
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/144 Training of HMMs

Abstract

For speech recognition a new word is represented as based on a stored inventory of models of sub-word units. First a plurality of utterances are presented that all should conform to the word. For building a word model from the utterances, these are represented by a sequence of feature vectors. First, the utterances are used to train a whole-word model that is independent of the models of the sub-word units. The length of the whole-word model equals the average length of the utterances. Next, a sequence of Markov states and associated probability densities of acoustic events of the whole-word model is interpreted as a reference template represented by a string of averaged feature vectors. Finally, the string is recognized by matching to models in the inventory and storing a recognition result as a model of the utterances.

Claims

1. A method for constructing a model of a new word for addition to a word model database of a speech recognition system which includes a plurality of pre-existing word models based on an inventory of models of sub-word units, the method comprising the steps of:

receiving a plurality of utterances that each purportedly conform to the new word;

determining an average length of the plurality of utterances;

constructing a whole word model of the new word, using the plurality of utterances but without reference to the inventory of models of sub-word units, wherein the whole word model has a length equal to the average length of the plurality of utterances;

constructing a reference template represented by a sequence of averaged feature values of the plurality of utterances; and

,

matching the sequence of averaged feature values to the models of sub-word units stored in the inventory of the speech recognition system in order to construct the model of the new word for addition to the word model database of the speech recognition system, whereby the thusly constructed model of the new word is based upon the inventory of models of sub-word units, thereby facilitating the recognition of subsequent utterances of the new word using the inventory of models of sub-word units.

2. The method as set forth in claim 1, wherein the constructing a reference template step is carried out by interpreting a sequence of Markov states and associated probability densities of acoustic events of the whole word model, and then developing the sequence of averaged feature values based upon the resultant interpretation.

3. A speech recognition device, comprising:

a word model database which includes a plurality of pre-existing word models based on an inventory of models of sub-word units;

means for receiving a plurality of utterances that each purportedly conform to a new word which does not correspond to any of the plurality of pre-existing word models included in the word model database;

means for determining an average length of the plurality of utterances;

means for constructing a whole word model of the new word, using the plurality of utterances, but without reference to the inventory of models of sub-word units, wherein the whole word model has a length equal to the average length of the plurality of utterances;

means for constructing a reference template represented by a sequence of averaged feature values of the plurality of utterances; and

,

means for matching the sequence of averaged feature values to the models of subword units stored in the inventory in order to construct the model of the new word for addition to the word model database, whereby the thusly constructed model of the new word is based upon the inventory of models of sub-word units, thereby facilitating the recognition of subsequent utterances of the new word using the inventory of models of sub-word units.

4. The speech recognition device as set forth in claim 3, wherein the means for constructing a reference template further includes means for interpreting a sequence of Markov states and associated probability densities of acoustic events of the whole word model, and then developing the sequence of averaged feature values based upon the resultant interpretation.