Method of speech recognition

• US 6,246,980 B1
• Filed: 09/28/1998
• Issued: 06/12/2001
• Est. Priority Date: 09/29/1997
• Status: Expired due to Term

• Current Assignee: RPX Clearinghouse LLC (RPX Corporation)
• Original Assignee: Nortel Networks Corporation
• Inventors: Lockwood, Philip, Lelievre, Laurent, Glorion, Catherine
• Primary Examiner(s): Tsang, Fan
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US09/161,588
• Time in Patent Office: 988 Days
• Field of Search: 704/231, 704/236, 704/239, 704/255, 704/256
• US Class Current: 704/231
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

A recognition test is applied to a segment of speech produced by a speaker. An examination is carried out to see whether a first optimum model, with which the recognition test has matched the spoken segment of speech, satisfies a criterion of likelihood. If the criterion of likelihood is satisfied, this first optimum model is selected. If not, a repetition of the speech segment is requested. The recognition test is applied to the segment of speech repeated by the speaker. This second optimum model is selected if it satisfies the likelihood criterion. If not, and if a combination of the results of the two recognition tests satisfies at least one criterion of combined selection, one of the models which the two recognition tests have matched with the spoken segment of speech and with the repeated segment is selected.

Claims

1. Method of speech recognition applying recognition tests to speech segments uttered by a speaker, wherein each recognition test applied to a speech segment provides a list of n sets of parameters memorised in a dictionary of references, n being a number at least equal to one, whereby the sets of said list present maximum likelihoods taking into account the observation of the speech segment, said list being ranked according to decreasing likelihood, the method comprising the steps of:

applying the recognition test to a first speech segment uttered by the speaker;

examining whether a first optimum set of parameters, obtained at the top of the list of n sets provided by the recognition test applied to the first speech segment, satisfies a criterion of likelihood;

if the criterion of likelihood is satisfied by the first optimum set of parameters, selecting the first optimum set of parameters;

if the criterion of likelihood is not satisfied by the first optimum set of parameters, requesting the speaker to repeat the speech segment;

applying the recognition test to a second speech segment as repeated by the speaker;

examining whether a second optimum set of parameters, obtained at the top of the list of n sets provided by the recognition test applied to the second speech segment, satisfies the criterion of likelihood;

if the criterion of likelihood is satisfied by the second optimum set of parameters, selecting the second optimum set of parameters;

if the criterion of likelihood is not satisfied by the second optimum set of parameters and if a combination of the results of the recognition test applied to the first and second speech segments

satisfies at least one criterion of combined selection, selecting a set of parameters included in both lists of n sets provided by the recognition test applied to the first and second speech segments,

wherein said at least one criterion of combined selection comprises the identity of the first and second optimum sets of parameters.

2. Method of learning according to claim 1, wherein the sets of parameters memorised in the dictionary of references define hidden Markov models.

3. Method of speech recognition applying recognition tests to speech segments uttered by a speaker, wherein each recognition test applied to a speech segment provides a list of n sets of parameters memorized in a dictionary of references, n being a number at least equal to two, whereby the sets of said list present maximum likelihoods taking into account the observation of the speech segment, said list being ranked according to decreasing likelihood, wherein the dictionary of references contains at least one set of rejection parameters, the method comprising the steps of:

applying the recognition test to a first speech segment uttered by the speaker;

examining whether a first optimum set of parameters, obtained at the top of the list of n sets provided by the recognition test applied to the first speech segment, satisfies a criterion of likelihood;

if the criterion of likelihood is satisfied by the first optimum set of parameters, selecting the first optimum set of parameters;

if the criterion of likelihood is not satisfied by the first optimum set of parameters, requesting the speaker to repeat the speech segment;

applying the recognition test to a second speech segment as repeated by the speaker;

examining whether a second optimum set of parameters, obtained at the to of the list of n sets provided by the recognition test applied to the second speech segment, satisfies the criterion of likelihood;

if the criterion of likelihood is satisfied by the second optimum set of parameters, selecting the second optimum set of parameters;

if the criterion of likelihood is not satisfied by the second optimum set of parameters and if a combination of the results of the recognition test applied to the first and second speech segments satisfies at least one criterion of combined selection, selecting a set of parameters included in both lists of n sets provided by the recognition test applied to the first and second speech segments,

wherein said at least one criterion of combined selection comprises both the appearance of a set of parameters, other than a set of rejection parameters, at the top of the list provided by the recognition test applied to one of the first and second speech segments, and the appearance of the same set of parameters in second position, after a set of rejection parameters, in the list provided by the recognition test applied to the other one of the first and second speech segments.

4. Method of learning according to claim 3, wherein the sets of parameters memorised in the dictionary of references define hidden Markov models.