US 7,689,414 B2 · Speech recognition device and method

Abstract

In a speech recognition device (1) for recognizing text information (TI) corresponding to speech information (SI), wherein speech information (SI) can be characterized in respect of language properties, there are firstly provided at least two language-property recognition means (20, 21, 22, 23), each of the language-property recognition means (20, 21, 22, 23) being arranged, by using the speech information (SI), to recognize a language property assigned to said means and to generate property information (ASI, LI, SGI, CI) representing the language property that is recognized, and secondly there are provided speech recognition means (24) that, while continuously taking into account the at least two items of property information (ASI, LI, SGI, CI), are arranged to recognize the text information (TI) corresponding to the speech information (SI).

Drawings

Drawings are temporarily unavailable. Open the PDF at USPTO

Drawings

Claims15

Dependent claims

A system for providing transcription of a conference between a plurality of participants of the conference, the system comprising:

a plurality of reception stages to receive information from the plurality of participants over a respective plurality of transmission channels;
and
at least one processor with programmed to receive the information from the plurality of reception stages, the at least one processor further programmed to:
analyze the information received at the plurality of reception stages to determine which of the plurality of participants of the conference is speaking during a given time interval based, at least in part, on identifying which of the plurality of reception stages is receiving speech information;
select one of the plurality of transmission channels corresponding to the reception stage identified as receiving speech information as an in-use channel;
determine channel information including at least one transmission parameter of the in-use channel;
extract at least one feature vector from the speech information based, at least in part, on the channel information;
perform acoustic segmentation of the speech information to generate acoustic segmentation information indicating at least one segment identified in the speech information based, at least in part, on the channel information and the at least one feature vector, the acoustic segmentation information including a label for the at least one segment of the speech information indicating whether the at least one segment is associated with speech, a pause in speech or non-speech;
determine a language of the speech information based, at least in part, on the channel information, the at least one feature vector and the acoustic segmentation information;
and
generate text information corresponding to words recognized in the speech information based, at least in part, on the channel information, the at least one feature vector, the acoustic segmentation information and the language.

8 more elements

A method of providing transcription of a conference between a plurality of participants of the conference, the method comprising:

receiving information over a plurality of transmission channels from the plurality of participants;
using at least one processor to analyze the information received at the plurality of reception stages to determine which of the plurality of participants of the conference is speaking during a given time interval based, at least in part, on identifying which of the plurality of reception stages is receiving speech information;
selecting one of the plurality of transmission channels corresponding to the reception stage identified as receiving speech information as an in-use channel;
determining channel information including at least one transmission parameter that identifies the in-use channel;
extracting at least one feature vector from the speech information based, at least in part, on the channel information;
performing acoustic segmentation of the speech information to generate acoustic segmentation information indicating at least one segment identified in the speech information based, at least in part, on the channel information and the at least one feature vector, the acoustic segmentation information including a label for the at least one segment of the speech information indicating whether the at least one segment is associated with speech, a pause in speech or non-speech;
determining a language of the speech information based, at least in part, on the channel information, the at least one feature vector and the acoustic segmentation information;
and
generating text information corresponding to words recognized in the speech information based, at least in part, on the channel information, the at least one feature vector, the acoustic segmentation information and the language of the speech information.

6 more elements

A computer readable storage device encoded with a plurality of instructions for execution on at least one processor, the plurality of instructions, when executed on the at least one processor, performing a method of providing transcription of a conference between a plurality of participants of the conference, the method comprising:

receiving information over a plurality of transmission channels from the plurality of participants;
analyzing the information received at the plurality of reception stages to determine which of the plurality of participants of the conference is speaking during a given time interval based, at least in part, on identifying which of the plurality of reception stages is receiving speech information;
selecting one of the plurality of transmission channels corresponding to the reception stage identified as receiving speech information as an in-use channel;
determining channel information including at least one transmission parameter that identifies the in-use channel;
extracting at least one feature vector from the speech information based, at least in part, on the channel information;
performing acoustic segmentation of the speech information to generate acoustic segmentation information indicating at least one segment identified in the speech information based, at least in part, on the channel information and the at least one feature vector, the acoustic segmentation information including a label for the at least one segment of the speech information indicating whether the at least one segment is associated with speech, a pause in speech or non-speech;
determining a language of the speech information based, at least in part, on the channel information, the at least one feature vector and the acoustic segmentation information;
and
generating text information corresponding to words recognized in the speech information based, at least in part, on the channel information, the at least one feature vector, the acoustic segmentation information and the language of the speech information.

6 more elements

Dependencies are inferred from the claim text.

Description16,940 words

Summary1,727 words

#The invention relates to a speech recognition device for recognizing text information corresponding to speech information.

#The invention further relates to a speech recognition method for recognizing text information corresponding to speech information.

#The invention further relates to a computer program product that is arranged to recognize text information corresponding to speech information.

Brief description of the drawings16 figures

#In the drawings:

#FIG. 1 is a schematic view in the form of a block circuit diagram of a speech recognition device according to one embodiment of the invention,

#FIG. 2 shows, in a similar way to FIG. 1, audio preprocessor means of the speech recognition device shown in FIG. 1,

Detailed description14,827 words

#Shown in FIG. 1 is a speech recognition device 1 that is arranged to recognize text information TI corresponding to speech information TI, and that forms a conference transcription device by means of which the speech information SI that occurs at a conference and is produced by conference participants when they speak can be transcribed into text information TI.

#The speech recognition device 1 is implemented in the form of a computer 1A, of which only the functional assemblies relevant to the speech recognition device 1 are shown in FIG. 1. The computer 1A has a processing unit that is not shown in FIG. 1 and an internal memory 1B, although only the functions of the internal memory 1B that are relevant to the speech recognition device 1 will be considered in detail below in connection with FIG. 1. The speech recognition device 1 uses the internal memory 1B to recognize the text information 1B corresponding to the speech information S1. The computer runs a computer program product that can be loaded directly into the memory 1B of the computer 1A and that has sections of software code.

#The speech recognition device 1 has reception means 2 that are arranged to receive speech information SI and to generate and emit audio signals AS representing the speech information SI, an audio signal AS bandwidth that affects the recognition of the speech information SI being dependent on a reception channel or transmission channel that is used to receive the speech information SI. The reception means 2 have a first reception stage 3 that forms a first reception channel and by means of which the speech information SI can be received via a plurality of microphones 4, each microphone 4 being assigned to one of the conference participants present in a conference room, by whom the speech information SI can be generated. Associated with the microphones 4 is a so-called sound card (not shown in FIG. 1) belonging to the computer 1A, by means of which the analog audio signals AS can be converted into digital audio signals AS. The reception means 2 also have a second reception stage 5 that forms a second reception channel and by means of which the speech information SI can be received via a plurality of analog telephone lines. The reception means 2 also have a third reception stage 6 that forms a third reception channel and by means of which the speech information SI can be received via a plurality of ISDN telephone lines. The reception means 2 also have a fourth reception stage 7 that forms a fourth reception channel and by means of which the speech information SI can be received via a computer data network by means of a so-called “voice-over-IP” data stream. The reception means 2 are also arranged to emit a digital representation of the audio signal AS received, in the form of a data stream, the digital representation of the audio signal AS having audio-signal formatting corresponding to the given reception channel and the data stream having so-called audio blocks and so-called audio headers contained in the audio blocks, which audio headers specify the particular audio-signal formatting.

Facts

Application

10/535,295Filed 31 Oct 2003

Priority

EP 0210262622 Nov 2002

Published

US 2006/0074667 A16 Apr 2006

Granted

30 Mar 2010

Term ended

7 Dec 2025

Estimated · how

20 years from earliest known nonprovisional domestic filing plus recorded patent term adjustment, including 768 days of adjustment. Maintenance-fee lapses, terminal disclaimers and statutory extensions have not been checked. The continuity chain may be incomplete; this is not an enforceability determination.

Classification6 codes

G10L 15/26
Speech recognition: speech to text systems
G10L 15/18
Speech classification or search: using natural language modelling
G10L 15/183
Using natural language modelling: using context dependencies, e.g. language models
G10L 15/22
Speech recognition: procedures used during a speech recognition process, e.g. man-machine dialogue

Timeline

22 Nov 2002Priority claimedEP 02102626
31 Oct 2003Application filedUS 10/535,295
17 May 2005Assigned to Koninklijke Philips Electronics, N.V.assignment of assignor's interest
17 May 2005Assigned to Koninklijke Philips Electronics, N.V.assignment of assignor's interest
6 Apr 2006PublishedUS 2006/0074667 A1
24 Feb 2009Assigned to Nuance Communications Austria GmbHassignment of assignor's interest
30 Mar 2010GrantedUS 7,689,414 B2
14 May 2013Assigned to Nuance Communications, Inc.assignment of assignor's interest
8 Nov 2023Assigned to Microsoft Technology Licensing, LLCassignment of assignor's interest
9 Nov 2023Assigned to Microsoft Technology Licensing, LLCassignment of assignor's interest
7 Dec 2025Term endedEstimated, includes 768 days of adjustment

Speech recognition device and method

Life of this patent

Speech recognition device and method

Life of this patent

Life of this patent

Abstract

Drawings

Drawings

Claims15

Description16,940 words

Drawings

Classification6 codes

Life of this patent

Description16,940 words