System and method of improving speech recognition using context

Patent metadata

Field	Value
Publication number	US-9626963-B2
Application number	US-201313874304-A
Country	US
Kind code	B2
Filing date	Apr 30, 2013
Priority date	Apr 30, 2013
Publication date	Apr 18, 2017
Grant date	Apr 18, 2017

Abstract

A system and method are provided for improving speech recognition accuracy. Contextual information about user speech may be received, and then speech recognition analysis can be performed on the user speech using the contextual information. This allows the system and method to improve accuracy when performing tasks like searching and navigating using speech recognition.

First claim

The invention claimed is:

1. A system, comprising:
a processor;
a single microphone configured to both record user speech and to record ambient sounds; and
a speech recognition module configured to:
identify that the ambient sounds are of a particular type by comparing the ambient sounds to stored waveforms;
select a dictionary based on the identified particular type of ambient sounds;
identify, as contextual information, terms related to the identified particular type of ambient sounds based on identification of the identified particular type of ambient sounds, the terms being generated as contextual information;
alter, in response to identification of the terms related to the identified particular type of ambient sounds, the dictionary such that the dictionary includes the terms related to the identified particular type of ambient sounds;
assign, in the dictionary, score values to the terms related to the identified particular type of ambient sounds based on identifying that the terms are related to the identified particular type of ambient sounds; and
analyze the user speech by comparing each potential output word or phoneme in the user speech to waveforms stored for the dictionary to attempt to match the potential output word or phoneme to a waveform corresponding to a particular word or phoneme in the dictionary, an analysis varying based on the assigned scores to the terms identified as contextual information.

2. The system of claim 1 , wherein the ambient sounds include music and the identification that the ambient sounds are of the particular type includes identifying that the ambient sounds are music and identifying the music, wherein the speech recognition module is further configured to retrieve identify, as the contextual information, terms related to the identified music.

3. The system of claim 1 , further comprising a sensor, and wherein the contextual information includes information identified from sensor information detected by the sensor.

4. The system of claim 3 , wherein the sensor is a global positioning system module and the contextual information includes location.

5. The system of claim 3 , wherein the sensor is a global positioning system module and the contextual information includes speed.

6. A method comprising:
recording sounds using a single microphone;
identifying, using one or more processors, potential output words and phonemes as well as ambient sounds in the sounds recorded by the single microphone;
identifying that the ambient sounds are of a particular type by comparing the ambient sounds to stored waveforms;
selecting a dictionary based on the identified particular type of ambient sounds;
identifying, as contextual information, terms related to the identified particular type of ambient sounds based on identification of the identified particular type of ambient sounds, the terms being generated as contextual information;
assigning, in the dictionary, score values to the terms related to the identified particular type of ambient sounds based on identifying that the terms are related to the identified particular type of ambient sounds; and
analyzing user speech by comparing each potential output word or phoneme in the user speech to waveforms stored for the dictionary to attempt to match the potential output word or phoneme to a waveform corresponding to a particular word or phoneme in the dictionary, the analyzing varying based on the assigned scores to the terms identified as contextual information.

7. The method of claim 6 , wherein the contextual information includes user location.

8. The method of claim 6 , wherein the contextual information includes speed of movement of a user.

9. The method of claim 6 , wherein the ambient sounds include music and the identification that the ambient sounds are of the particular type includes identifying that the ambient sounds are music and identifying the music, wherein the speech recognition module is further configured to identify, as the contextual information, terms related to the identified music.

10. The method of claim 6 , further comprising altering the dictionary based on the contextual information such that the dictionary includes the terms related to the identified particular type of ambient sounds.

11. The method of claim 10 , wherein the dictionary is altered by replacing the dictionary with a different dictionary.

12. The method of claim 10 , wherein the dictionary is altered by adding words pertaining to the contextual information to the dictionary.

13. A non-transitory machine-readable storage medium comprising a set of instructions which, when executed by a processor, causes execution of operations comprising:
recording sounds using a single microphone;
identifying potential output words and phonemes as well as ambient sounds in the sounds recorded by the single microphone;
identifying that the ambient sounds are of a particular type by comparing the ambient sounds to stored waveforms;
selecting a dictionary based on the identified particular type of ambient sounds;
identifying, as contextual information, terms related to the identified particular type of ambient sounds based on identification of the identified particular type of ambient sounds, the terms being generated as contextual information;
assigning, in the dictionary, score values to the terms related to the identified particular type of ambient sounds based on identifying that the terms are related to the identified particular type of ambient sounds; and
analyzing the user speech by comparing each potential output word or phoneme in the user speech to waveforms stored for the dictionary to attempt to match the potential output word or phoneme to a waveform corresponding to a particular word or phoneme in the dictionary, the analyzing varying based on the assigned scores to the terms identified as contextual information.

14. The non-transitory machine-readable storage medium of claim 13 , wherein the speech recognition analysis includes utilizing a hidden Markov model.

15. The non-transitory machine-readable storage medium of claim 13 , wherein the ambient sounds include music and the identification that the ambient sounds are of the particular type includes identifying that the ambient sounds are music and identifying the music, wherein the speech recognition module is further configured to identify, as the contextual information, terms related to the identified music.

Assignees

• Paypal Inc

Inventors

• Farraro Eric J

Classifications

• G10L15/06

  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 title

• G10L15/063 Primary

  Training · CPC title

• G10L15/30 Primary

  Distributed recognition, e.g. in client-server systems, for mobile phones or network applications · CPC title

• G10L2015/226

  using non-speech characteristics · CPC title

• G10L15/22 Primary

  Procedures used during a speech recognition process, e.g. man-machine dialogue · CPC title