Method of supporting the proof-reading of speech-recognized text with a replay speed adapted to the recognition reliability

• US 7,480,613 B2
• Filed: 07/31/2002
• Issued: 01/20/2009
• Est. Priority Date: 08/04/2001
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Inventors: Kellner, Andreas
• Primary Examiner(s): Hudspeth; David R
• Assistant Examiner(s): Serrou; Abdelali
• Application Number:
  US10/210,835
• Publication Number:
  US 20030028375A1
• Time in Patent Office: 2,365 Days
• Field of Search: 704/231, 704234-235, 704/251, 704/211
• US Class Current: 704/231
• CPC Class Codes: G10L 13/00 Speech synthesis; Text to speech systems
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2015/225 Feedback of the input speech

Abstract

The invention relates to a method of supporting the proof-reading of a text (2, 30) obtained in particular by speech recognition from a speech signal (1), of which at least one text component (B1 to B5) has a reliability level (3) for its correctness, which method comprises a control of the progression of the replay speed (12, 13) of the part of the speech signal (1) belonging to the text component (B1 to B5) as a function of the reliability level (3) of the text component (B1 to B5).

Faster replay of the reliable passages of a speech signal (1) shortens the working time required for proof-reading, while the slow replay of the unreliable passages raises the attention of the proof-reading user. An acoustic and/or visual (32) marking of the passage as a function of its reliability (3) is advantageous. The method according to the invention is particularly suitable for implementation within a speech recognition system.

Claims

1. A computer-implemented method for supporting proof-reading of a text obtained by speech recognition of a speech signal, said method comprising:

dividing said text into components having respective reliability levels for correctness of their speech recognition, wherein a rate of speech replay of each of said components is based on the associated reliability level; and

,

in said proof-reading continuously throughout said text, controlling progression of replay speed of the speech signal as a function of the respective reliability levels of the text components, wherein the replay speed is adjusted based on the rate of speech replay of temporally adjacent components to allow a smooth transition from one component replay speed to another component replay speed.

4. A method as claimed in claim 1, wherein a value range of the reliability levels is divided into sub-ranges, and in that the progression of the replay speed is controlled as a function of the sub-range to which the reliability level of the text component is allocated.

13. The method of claim 1, wherein at least three of said reliability levels mutually differ.

15. The device of claim 1, wherein said progression involves respective replay speeds for said at least three reliability levels.

2. A method of supporting the proof-reading of a text obtained by speech recognition of a speech signal, of which at least one text component has a reliability level for the correctness of its speech recognition, wherein the method comprises the steps of:

control of the progression of the replay speed of the part of the speech signal belonging to the text component as a function of the reliability level of the text component, wherein the progressions of the replay speeds of temporally adjacent parts of the speech signal are adjusted to each other to allow a smooth transition from one component replay speed to another component replay speed.

3. A method as claimed in claim 2, wherein the progression of the replay speed of the part of the speech signal is controlled without changing the gradient of the pitch of the part of the speech signal.

5. A method as claimed in claim 2, wherein the reliability level of the text component is marked acoustically and/or visually.

6. A method as claimed in claim 2, further comprising:

providing a user of the method with the capability to configure the dependency between the progression of the replay speed and the reliability level of the text component and/or the degree of adjustment of the progressions of the replay speeds of temporally adjacent parts of the speech signal and/or the sub-ranges of the reliability levels and/or the method of acoustic and/or visual marking of the reliability level of the text component.

7. A device for supporting the proof-reading of a text obtained by speech recognition of a speech signal, of which at least one text component has a reliability level for the correctness of its speech recognition, wherein the device is designed to control the progression of the replay speed of the part of the speech signal (1) belonging to the text component as a function of the reliability level of the text component, wherein the progressions of the replay speeds of temporally adjacent parts of the speech signal are adjusted to each other to allow a smooth transition from one component replay speed to another component replay speed.

14. The device of claim 7, wherein at least three reliability levels of respective text components mutually differ.

8. A speech recognition system with a device for supporting proof-reading that is continuous throughout a text obtained by speech recognition of a speech signal, said text being divided into text components having respective reliability levels for the correctness of their speech recognition, wherein the device is designed to control, during the continuous proof-reading, progression of replay speed of the speech signal as a function of the reliability level of the text component adjusted to allow a smooth transition from one component replay speed to another component replay speed.

16. The system of claim 8, wherein at least three of said reliability levels mutually differ.

17. The system of claim 16, wherein said progression involves respective, different replay speeds for said reliability levels.

9. A computer-implemented method of supporting proof-reading that is continuous throughout a text obtained by transcription into written form of a speech signal and divided into respective text components having respective reliability levels for their correctness, respective parts of the speech signal belonging to the text components being known, wherein said method includes:

deriving the reliability levels; and

controlling, during the continuous proof-reading, progression of replay speed of the speech signal as a function of the reliability levels of the respective text components to allow a smooth transition from one component replay speed to another component replay speed.

18. The method of claim 9, wherein said transcription is manual or obtained by manual correction of speech-recognized text.

19. The method of claim 9, wherein said text being proof-read was written by a writer, said deriving comprises producing error statistics from previous corrections by the writer.

10. A computer-implemented method for supporting proof-reading of a text obtained by transcription into written form of a speech signal, comprising:

determining, throughout said text, which part of the speech signal belongs to which component of said text by time warping the text to the speech signal;

determining a reliability level for correctness of the text component, controlling progression of replay speed of the part of the speech signal belonging to the text component as a function of the reliability level of the text component; and

adjusting said replay rate to allow a smooth transition from one component replay speed to another component replay speed.

11. The method of claim 10, wherein said reliability level is a variable having more than two values within a domain of said function.

12. The method of claim 11, wherein said progression involves respective replay speeds for said more than two values.