Select a recognition error by comparing the phonetic

• US 20030061043A1
• Filed: 09/13/2002
• Published: 03/27/2003
• Est. Priority Date: 09/17/2001
• Status: Active Grant

• Current Assignee: Nuance Communications Austria Gmbh (Microsoft Corporation)
• Original Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Inventors: Gschwendtner, Wolfgang
• Granted Patent: US 6,735,565 B2
• Time in Patent Office: Days
• US Class Current: 704/254
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 2015/221 Announcement of recognition results
• CPC Class Codes: G10L 2015/225 Feedback of the input speech

Abstract

A correction device (4) for a speech recognition device (2) is provided, with which the replacement of incorrectly recognized words (FETI) of the recognized text (ETI) is especially simple to execute. The correction device (4) is based on the recognition that the phoneme sequences of incorrectly recognized words and the spoken words actually to be recognized are very similar, and automatically marks words in the recognized text (ETI) which show a phoneme sequence similar to that of a correction word (KWI) put in by the user.

Claims

1. A correction device (4) for correcting a text (ETI) recognized by a speech recognition device (2) for a spoken text (GTI), where the recognized text (ETI) for spoken words of the spoken text (GTI) includes correctly recognized words and incorrectly recognized words (FETI), with

input means (13) for receiving at least one manually input correction word (KWI), in order to replace at least one of the incorrectly recognized words (FETI) with the at least one correction word (KWI), and with

transcription means (16) for phonetically transcribing at least the input correction word (KWI) into a phoneme sequence (PI(KWI)) and with

search means (17) for finding the phoneme sequence (PI(KWI)) of the at least one correction word (KWI) in phoneme sequences (PI(KTI)) of the words of the recognized text and for issuing position information (PI) which identifies the position of at least one word within the recognized text (ETI) whose phoneme sequence essentially matches the phoneme sequence (PI(KWI)) of the at least one correction word (KWI), and with

output means (17) for issuing said position information (PI) so as to enable an marking of the at least one word identified by the position information (PI) in the recognized text information (ETI).

2. A correction device (4) as claimed in claim 1, wherein the correction device (4) is designed for interrupting a synchronous reproduction, in which the spoken words (GTI) of the spoken text are acoustically reproduced and the recognized words of the recognized text (ETI) for the spoken words (GTI) are synchronously optically marked, when a correction word

3. A correction device (4) as claimed in claim 2, wherein the correction device (4) is designed for terminating the interruption of the synchronous reproduction when the replacement of the at least one word identified by the position information (PI) with the at least one correction word (KWI) has been confirmed by manual input of a confirmation.

4. A correction device (4) as claimed in claim 2, wherein the search means (17) are designed so as to search for the phoneme sequence (PI(KWI)) of the at least one correction word (KWI) in the phoneme sequences (PI(ETI)) of the words contained in a search area of the recognized text, said search area being defined by a number of M words before and N words behind the last marked word in the recognized text (ETI) before the interruption of the synchronous reproduction.

5. A correction device (4) as claimed in claim 1, wherein the search means (17) are designed so as to search for the phoneme sequence (PI(KWI)) of the at least one correction word (KWI) in phoneme sequences determined by the speech recognition device (2) from the spoken words of the spoken text (GTI).

6. A correction device (4) as claimed in claim 5, wherein the correction device (4) is designed so as to form part of the speech recognition device (2).

7. A correction device (4) as claimed in claim 1, wherein the search means (17) are designed for ignoring phonemes that differ from the compared phoneme sequences but sound similar in the search for essentially matching phoneme sequences.

8. A correction method for correcting a text (GTI) recognized by a speech recognition device (2) for a spoken text, the recognized text (ETI) for spoken words of the spoken text (GTI) including correctly recognized words and incorrectly recognized words (FETI), the following steps being processed:

receiving at least one manually entered correction word (KWI), so as to replace at least one of the incorrectly recognized words (FETI) with the at least one correction word (KWI);

phonetically transcribing at least the input correction word (KWI) into a phoneme sequence (PI(KWI));

searching for the phoneme sequence of the at least one correction word (KWI) in phoneme sequences (PI(ETI)) of the words of the recognized text (ETI) and issuing position information (PI) which identifies the position of at least one word within the recognized text (ETI) whose phoneme sequence essentially matches the phoneme sequence of the at least one correction word (KWI);

issuing the position information (PI) so as to enable marking of the at least one word identified by the position information (PI) in the recognized text information (ETI).

9. A correction method as claimed in claim 8, wherein the following further process step is performed:

interrupting a synchronous reproduction, in which the spoken words of the spoken text (GTI) are acoustically reproduced and the recognized words of the recognized text (ETI) are synchronously optically marked for the spoken words (GTI), when a correction word (KWI) is manually put in.

10. A correction method as claimed in claim 9, wherein the following further process step is performed:

terminating the interruption of the synchronous reproduction when the replacement of the at least one word identified by the position information (PI) with the at least one correction word (KWI) has been confirmed by manual input of a confirmation.

11. A correction method as claimed in claim 9, wherein the following further process step is performed:

searching for the phoneme sequence of the at least one correction word (KWI) in the phoneme sequences of the words contained in a search area of the recognized text (ETI), the search area being defined by a number of M words before and N words behind the last marked word in the recognized text (ETI) before the interruption of the synchronous reproduction.

12. A correction method as claimed in claim 8, wherein the following further process step is performed:

searching for the phoneme sequence (PI(KWI)) of the at least one correction word (KWI) in phoneme sequences determined by the speech recognition device (2) from the spoken words of the spoken text (GTI).

13. A correction method as claimed in claim 8, wherein the following further process step is performed:

searching for essentially matching phoneme sequences, the phonemes that differ from the compared phoneme sequences but sound similar being ignored.