Method to train the language model of a speech recognition system to convert and index voicemails on a search engine

• US 7,415,409 B2
• Filed: 12/03/2007
• Issued: 08/19/2008
• Est. Priority Date: 12/01/2006
• Status: Active Grant

• Current Assignee: Coveo Solutions, Inc.
• Original Assignee: Coveo Solutions, Inc.
• Inventors: Simoneau, Laurent, Soucy, Pascal
• Primary Examiner(s): Lerner; Martin
• Application Number:
  US11/949,346
• Publication Number:
  US 20080133235A1
• Time in Patent Office: 260 Days
• Field of Search: 704/10, 704/231, 704/243, 704/244, 704/255, 704/256.3, 379/88.01, 379/88.04
• US Class Current: 704/243
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams
• CPC Class Codes: H04M 2201/405 involving speaker-dependent recognition adaptation to speaker for speech recognition G10L15/07
• CPC Class Codes: H04M 3/4936 Speech interaction details speech recognition per se G10L15/00
• CPC Class Codes: H04M 3/53333 Message receiving aspects

Abstract

A method and a related system to index voicemail documents by training a language model for a speaker or group of speakers by using existing emails and contact information on available repositories.

Claims

1. A method for training a language model of a speech recognition engine, the method comprising:

extracting textual content from email documents from an email document source;

extracting contact data from a contact from a contact source comprising contacts, the contact data comprising at least one of a person'"'"'s name and email address;

forming a training set comprising the email documents, the email documents each having recipient or sender information that comprises the at least one of a person'"'"'s name and email address, the forming a training set further comprises;

extracting metadata from the email documents;

providing manually created sentence templates;

creating new training sentences by filling the sentence templates using the metadata; and

adding the new training sentences to the training set; and

training the language model using the textual content from the email documents and the new training sentences in the training set to produce a speech recognition profile for the contact.

2. The method of claim 1, further comprising grouping contacts into groups, each group comprising at least one contact;

generating, for each group of contacts, the training set using the contact data for the contacts in the group and using the textual content of the email documents whose recipient or sender contains a string that at least partially matches a person'"'"'s name and email address; and

the training comprises training the language model using the training set.

3. The method of claim 1, further comprising:

extracting audio content for a given voicemail from a voicemail source;

identifying the contact associated to a person leaving the given voicemail;

selecting the speech recognition profile that corresponds to the identified contact;

converting said audio content of the voicemail into a transcription using the selected speech recognition profile; and

indexing the transcriptions thereby resulting in an indexing of the voicemails.

4. The method of claim 3 wherein said identifying the contact of the person leaving the given voicemail is conducted by obtaining identification information of the device from which the voicemail originates and retrieving the contact that contains the identification information.