Calibration of a speech recognition engine using validated text

• US 9,218,807 B2
• Filed: 01/07/2011
• Issued: 12/22/2015
• Est. Priority Date: 01/08/2010
• Status: Active Grant

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Hon-Anderson, Eric, Stuller, Robert W.
• Primary Examiner(s): He, Jialong
• Application Number:
  US12/986,855
• Publication Number:
  US 20110301940A1
• Time in Patent Office: 1,810 Days
• Field of Search: 704/232, 704/243, 704/259, 704/275
• US Class Current: 1/1
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks
• CPC Class Codes: G10L 2015/0638 Interactive procedures

Abstract

A system and method provide acoustic training of a voice or speech recognition engine and/or voice or speech recognition software application. Instead of requiring a user to read from a prepared or predetermined script, the system and method described herein enable acoustic training using any free text spoken phrases provided by the user directly, or by a previously recorded speech, presentation, or the like, performed by the user.

Claims

1. A speech recognition system that can be acoustically trained with free text audio, the system comprising:

a speech recognition software application operating on a computing device having a processor, the speech recognition software application comprising;

a speech recognition engine configured to receive the free text audio at the speech recognition engine which is unknown to the speech recognition engine previous to acoustical training of the speech recognition engine in both spoken audio and text forms, translate the free text audio into text form for display to a user, and receive a reviewed version of the text form and convert the reviewed version of the text form into a context free grammar based on text indicated as validated text as indicated by the user;

a comparison module configured to receive an indication of the validated text and associate the validated text with at least one word from the free text audio; and

a plurality of voice models;

wherein upon receipt of a plurality of instances in which validated text is associated with the at least one word from the free text audio, the speech recognition software application selects a subset of voice models of the plurality of voice models in such a way that the subset of voice models shares a plurality of characteristics with the free text audio associated with the validated text.

2. The system of claim 1, wherein the speech recognition software application is configured to record each instance of validated text.

3. The system of claim 1, wherein the speech recognition software application is configured to record each instance of validated text, accumulating instances of validated text up to a first predetermined number of instances of validated text or duration of audio signal, and further wherein the speech recognition software application is configured to perform calibration of the speech recognition engine once the first predetermined number of instances of validated text or duration of audio signal has been achieved.

4. The system of claim 3, wherein calibration of the speech recognition engine comprises the speech recognition engine selecting initial properties of an acoustic match to a voice model.

5. The system of claim 1, wherein the speech recognition software application is configured to record each instance of validated text, accumulate instances of validated text up to a second predetermined number of instances of validated text or duration of audio signal, and further wherein the speech recognition software application is configured to perform refining calibration of the speech recognition engine once the second predetermined number of instances of validated text or duration of audio signal has been achieved.

6. The system of claim 1, wherein the free text audio comprises a previously recorded audio recording of the user'"'"'s voice speaking.

7. The system of claim 1, wherein the free text audio comprises a real-time data representation of the user'"'"'s voice speaking.