Method, system and apparatus for improved voice recognition

• US 7,689,417 B2
• Filed: 09/04/2006
• Issued: 03/30/2010
• Est. Priority Date: 09/04/2006
• Status: Active Grant

• Current Assignee: Fortemedia, Inc.
• Original Assignee: Fortemedia, Inc.
• Inventors: Zhang, Ming, Hou, Zhen, Bai, Bo-Ren, Huang, Yen-Son Paul, Yu, Hang, Liu, Yaying
• Primary Examiner(s): ABEBE, DANIEL DEMELASH
• Application Number:
  US11/469,893
• Publication Number:
  US 20080059191A1
• Time in Patent Office: 1,303 Days
• Field of Search: 704/246, 704/260, 704/275, 379/88.02
• US Class Current: 704/246
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: G10L 2015/228 of application context

Abstract

An improved voice recognition system in which a Voice Keyword Table is generated and downloaded from a set-up device to a voice recognition device. The VKT includes visual form data, spoken form data, phonetic format data, and an entry corresponding to a keyword, and TTS-generated voice prompts and voice models corresponding to the phonetic format data. A voice recognition system on the voice recognition device is updated by the set-up device. Furthermore, voice models in the voice recognition device are modified by the set-up device.

Claims

1. A method for improved voice recognition in a system having a set-up device and a voice recognition device, comprising the steps of:

generating a Voice Keyword Table (VKT) and downloading the VKT to the voice recognition device;

upgrading a voice recognition system on the voice recognition device; and

modifying a voice model in the voice recognition device;

wherein the VKT comprise visual form data, spoken form data, phonetic format data, and an entry corresponding to a keyword, and TTS generated voice prompts and voice models corresponding to the phonetic format data.

2. The method of claim 1, wherein the step of generating the VKT comprises the steps of:

inputting visual form data and entry data;

transforming visual form data to default spoken form data;

mapping spoken form data to phonetic format; and

performing TTS-guided-pronunciation editing to modify phonetic format data.

3. The method of claim 2, further comprising the step of performing a confusion test using the phonetic format data, voice models and a confusion table to identify keywords in a confusion set.

4. The method of claim 3, further comprising the step of editing spoken form data and mapping edited spoken form data to phonetic format data to eliminate keywords from the confusion set.

5. The method of claim 3, further comprising the step of adapting keywords to eliminate keywords from the confusion set.

6. A method for improved voice recognition in a system having a set-up device and a voice recognition device, comprising the steps of:

generating a Voice Keyword Table (VKT) and downloading the VKT to the voice recognition device;

upgrading a voice recognition system on the voice recognition device; and

modifying a voice model in the voice recognition device;

wherein the step of modifying the voice model in the voice recognition device comprises the steps of;

building a keyword model from keywords in the VKT;

selecting keywords for adaptation;

obtaining new speech input for selected keywords;

adapting voice models for selected keywords using existing keyword voice models and new speech input to produce adapted voice models; and

downloading adapted speech models to the voice recognition device.

7. A method for improved voice recognition in a system having a set-up device and a voice recognition device, comprising the steps of:

generating a Voice Keyword Table (VKT) and downloading the VKT to the voice recognition device;

upgrading a voice recognition system on the voice recognition device; and

modifying a voice model in the voice recognition device;

wherein the step of modifying a voice model in the voice recognition device comprises the steps of;

downloading a new voice model from a network to the set-up device;

if the new voice model is a newer version than the voice model on the voice recognition device, determining if accumulated personal acoustic data exists;

if accumulated personal acoustic data exists, uploading the VKT from the voice recognition device to the set-up device, building a keyword model for adaptation from keywords in the uploaded VKT, performing adaptation using the new voice model and accumulated personal data to produce an adapted new voice model, and downloading the adapted new voice model to the voice recognition device; and

if no accumulated speech data exists, uploading the VKT to the set-up device, and building a keyword model for keywords in the uploaded VKT using the new voice model, and downloading the updated new voice model to the voice recognition device.

8. The method of claim 7, wherein the accumulated personal acoustic data is speech input recorded during user-initiated adaptation of voice models and stored on the set-up device.

9. The method of claim 7, wherein the accumulated personal acoustic data is speech input recorded during use of the voice recognition device to identify keywords and stored on the voice recognition device.

10. A method for improved voice recognition in a system having a set-up device and a voice recognition device, comprising the steps of:

generating a Voice Keyword Table (VKT) and downloading the VKT to the voice recognition device;

upgrading a voice recognition system on the voice recognition device;

modifying a voice model in the voice recognition device;

saving run-time information in the voice recognition device;

uploading saved run-time information from the voice recognition device to the set-up device;

processing run-time information on the set-up device; and

updating the voice recognition device according to the results of the processing of run-time information on the set-up device to improve voice recognition performance.

11. A method for improved voice recognition in a system having a set-un device and a voice recognition device, comprising the steps of:

generating a Voice Keyword Table (VKT) and downloading the VKT to the voice recognition device;

upgrading a voice recognition system on the voice recognition device;

modifying a voice model in the voice recognition device; and

initiating a diagnostic test on the voice recognition device by the set-up device.

12. The method of claim 11, wherein the diagnostic test further comprise a step of generating a user alert when voice recognition device hardware is not functioning properly.

13. The method of claim 11, wherein the diagnostic test further comprises a step of optimizing a microphone volume and a speaker volume on the voice recognition device.

14. The method of claim 11, wherein the step of initiating a diagnostic test further comprises a step of testing the connection between the set-up device and the voice recognition device.

15. A voice recognition system installed on a set-up device for improving voice recognition on a voice recognition device comprising:

a Voice Keyword Table (VKT) generating means for generating a VKT and downloading the VKT to the VKT to the voice recognition device; and

means for updating voice models on the voice recognition device;

wherein the VKT comprises visual form data, spoken form data, phonetic format data, and an entry corresponding to a keyword, and TTS-generated voice prompts and voice models corresponding to the phonetic format data.

16. The voice recognition system of claim 15, further comprising means for performing a confusion test to identify keywords in a confusion set.

17. A voice recognition system installed on a set-up device for improving voice recognition on a voice recognition device comprising:

a Voice Keyword Table (VKT) generating means for generating a VKT and downloading the VKT to the VKT to the voice recognition device;

means for updating voice models on the voice recognition device; and

means for updating the voice recognition device according to the results of the processing of run-time information saved on the voice recognition device to improve voice recognition performance.

18. A voice recognition system installed on a set-up device for improving voice recognition on a voice recognition device comprising:

a Voice Keyword Table (VKT) generating means for generating a VKT and downloading the VKT to the VKT to the voice recognition device;

means for updating voice models on the voice recognition device; and

means for initiating a diagnostic test on the voice recognition device.

19. An apparatus for improved voice recognition, comprising:

a set-up device comprising a first Voice Keyword Table (VKT) and a first voice recognition system; and

a voice recognition device comprising a second VKT corresponding to the first VKT and a second voice recognition system, the voice recognition device connectible to the set-up device through an interface;

wherein the first VKT and the second VKT comprise visual format data, spoken form data, phonetic format data, and an entry corresponding to a keyword, and TTS-generated voice prompts and voice models corresponding to the phonetic format data.