Apparatus for converting speech to text

• US 6,366,882 B1
• Filed: 03/27/1998
• Issued: 04/02/2002
• Est. Priority Date: 03/27/1997
• Status: Expired due to Term

• Current Assignee: Speech Machines, PLC (Lernout & Hauspie Investment Co. NV)
• Original Assignee: Speech Machines, PLC (Lernout & Hauspie Investment Co. NV)
• Inventors: Bijl, David, Hyde-Thomson, Henry
• Primary Examiner(s): Tsang, Fan
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US09/049,744
• Time in Patent Office: 1,467 Days
• Field of Search: 704/235, 704/245, 704/243, 704/244, 704/236, 704/270, 704/277
• US Class Current: 704/235
• CPC Class Codes: G06F 3/16 Sound input; Sound output speech processing G10L
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
• CPC Class Codes: H04M 2201/40 using speech recognition speech recognition per se G10L15/00
• CPC Class Codes: H04M 2201/60 Medium conversion
• CPC Class Codes: H04M 3/5307 for recording messages comprising any combination of audio and non-audio components

Abstract

A speech-to-text conversion system which comprises a plurality of user terminals for recording speeches, at least one automatic speech recognition processor to generate text from recorded speeches, and communication links to return corresponding texts to respective users. At least one automatic speech recognition processor improves recognition accuracy using data of the recorded speeches and the corresponding texts, the data being selected dependent upon subject matter area.

Claims

1. A transcription system for generating text for users in accordance with recorded speeches making use of automatic speech recognition, the system comprising:

a plurality of user terminals for recording speeches;

at least one automatic speech recognition processor to generate text from recorded speeches; and

communication means operative to return corresponding text to respective users;

wherein at least one automatic speech recognition processor is adapted to improve recognition accuracy for a user using selected data of the recorded speeches and the corresponding texts of at least one other user;

said speeches each having an identifier associated with it corresponding to subject matter area and/or user accent; and

said data being selected and pooled dependent upon identifiers common to said speeches;

said automatic speech recognition processor having a different language model for each identifier; and

each language model being adapted using pooled data with the corresponding identifier.

2. A system as claimed in claim 1, wherein each language model is adapted to acquire new words which occur in the corresponding subject matter area.

3. A system as claimed in claim 1, wherein the automatic speech recognition processor learns the probabilities of word occurrences dependent on subject matter area.

4. A system as claimed in claim 1, wherein upon the recorded speech being received by one of said automatic speech recognition processors, the identifier of user accent is used to select an acoustic model to be applied by the processors in automatic speech recognition and the identifier of subject matter area is used to select a language model to be applied in automatic speech recognition.

5. A system as claimed in claim 1, wherein said at leat one user terminal is remote from said at least one automatic speech recognition processor.

6. A system as claimed in claim 1, wherein the speech to text convertor includes a server remote from said at least one user terminal, the server being operative to control transfer of recorded speech files to a selected automatic speech recognition processor.

7. A system s claimed in claim 1, wherein each user terminal communicates the recorded speech files to the remote server by electronic mail.

8. A system as claimed in claim 1, wherein text files resulting from automatic speech recognition are sent to correction units.

9. A system as claimed in claim 8, wherein the correction units are remote from the automatic speech recognition processors.

10. A system as claimed in claim 8, wherein communications from the automatic speech recognition processors to each correction unit are undertaken under the control of the server.

11. A system as claimed in claim 8, wherein the correction units are remotely distributed.

12. A system as claimed in claim 8, wherein the correction units communicate to said at least one user terminal by electronic mail.

13. A system as claimed in claim 8, wherein each correction unit includes a visual display unit for display of the text and a manual interface usable to select text.

14. A system as claimed in claim 8, wherein correction is effected by a manual operation.

15. A system as claimed in claim 8, wherein corrections are recorded and transmitted back to the automatic speech recognition processor which undertook the automatic speech recognition for adaptation of the operation of the automatic speech recognition processor.

16. A system as claimed in claim 8, wherein corrections are sent by electronic mail.

17. A system as claimed in claim 1, wherein the recorded speech is continuous speech.

18. A system as claimed in claim 1, wherein a server acts to control assignment of recorded speech files for processing to automatic speech processors by queuing the received speech files and submitting them according to predetermined rules.

19. A system as claimed in claim 18, wherein the predetermined rules by which the server queues jobs is one of urgency and user priority rating.

20. A system as claimed in claim 1, wherein speech to text conversion is done as a single fully automatic operation.

21. A system as claimed in claim 1, wherein speech to text conversion is done as a part-automatic and part-manual operation using the automatic speech recognition processor and correction unit, respectively.

22. A transcription method for generating text for users in accordance with recorded speeches making use of automatic speech recognition, comprising:

recording speeches at a plurality of user terminals;

using at least one automatic speech recognition processor to generate text from said recorded speeches; and

returning corresponding text to respective users;

wherein

selected data of said recorded speeches and corresponding text of multiple users are used to adapt operation of at least one automatic speech recognition processor to improve recognition accuracy;

each speech being marked with an identifier corresponding to subject matter area and/or user accent;

said data being selected and pooled dependent upon identifiers common to said speeches;

a different language model being used by the automatic speech recognition processor for speeches having different identifier; and

each language model being adapted using pooled data with a corresponding identifiers.