Automatic speech analysis

• US 8,725,518 B2
• Filed: 04/25/2006
• Issued: 05/13/2014
• Est. Priority Date: 04/25/2006
• Status: Active Grant

• Current Assignee: Nice Ltd
• Original Assignee: Nice Systems Limited (Nice Ltd)
• Inventors: Waserblat, Moshe, Eilam, Barak
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): ADESANYA, OLUJIMI A
• Application Number:
  US11/411,764
• Publication Number:
  US 20070250318A1
• Time in Patent Office: 2,940 Days
• Field of Search: 704/236, 704/276
• US Class Current: 704/276
• CPC Class Codes: G10L 15/07 to the speaker
• CPC Class Codes: G10L 15/14 using statistical models, e.g. Hidden Markov Models [HMMs] G10L15/18 takes precedence

Abstract

A system for providing automatic quality management regarding a level of conformity to a specific accent, including, a recording system, a statistical model database with statistical models representing speech data of different levels of conformity to a specific accent, a speech analysis system, a quality management system. Wherein the recording system is adapted to record one or more samples of a speakers speech and provide it to the speech analysis system for analysis, and wherein the speech analysis system is adapted to provide a score of the speakers speech samples to the quality management system by analyzing the recorded speech samples relative to the statistical models in the statistical model database.

Claims

1. A system for providing quality management related to an accent of a call responder handling a call in a contact center, comprising:

a statistical model database with statistical accent models, each statistical accent model associated with an accent, based on accent related details from collections of speech instances that are further normalized relative to differences therebetween for accent conformity, and with a model accent score representing the conformity of the statistical accent model to the associated accent, and created from speech data;

a recording system for recordings a speech sample of a call responder having a call responder accent;

a speech analysis system for receiving said speech sample;

associating a score with the call responder accent by analyzing said speech sample relative to the statistical accent models in said statistical model database, and indicating a specific model of the statistical accent models to which the speech sample is closest, and automatically amending in said speech sample by replacement waveforms, sections determined as having known errors, whereby the amended speech sample replaces said speech sample to conform to the specific accent model; and

providing during the call conducted by the call responder an immediate visual feedback indicating the level of deviation of the agent from the required accent.

2. A system according to claim 1, wherein analysis of said speech sample by said speech analysis system comprises:

preparing a statistical model from said speech sample; and

comparing it to other statistical models in said statistical model database to determine the specific model and a level of conformity to the specific model.

3. A system according to claim 1, wherein said quality management system is adapted to provide feedback regarding the call responder accent based on a current score and a previous score of the call responder.

4. A system according to claim 1, wherein said feedback further comprises specific speech segments with a maximum level of deviation from a model associated with the call responder accent and with a highest score.

5. The system of claim 1, wherein said statistical model database represents people with different speech characteristics.

6. The system of claim 5, wherein the different speech characteristics are accents of a man, a woman or a child.

7. The system of claim 1, wherein the statistical accent models are created using at least one feature selected from the group consisting of:

phonology, acoustic phonetics, and phonotactics.

8. The system according to claim 1, further comprising a quality management system for providing feedback related to the accent score to a supervisor of the call responder.

9. The system according to claim 1, further comprising asking a caller to repeat a specific sentence with pre-selected sounds that are known to differentiate between different accents.

10. The system according to claim 1, further comprising assigning the call responder with an accent score corresponding to the model accent score associated with the specific model.

11. A method for providing quality management related to an accent of a call responder handling a call in a contact center, comprising:

creating one or more statistical accent models from one or more training speech data, each statistical accent model associated with an accent, based on accent related details from collections of speech instances that are further normalized relative to differences therebetween for accent conformity, and with a model accent score representing the conformity of the statistical accent model to the associated accent;

recording by a recording system a speech sample of a call responder having a call responder accent;

inputting the speech sample for analysis;

preparing a statistical speech model representing the speech sample of the call responder;

comparing the statistical speech model with said statistical accent models;

determining a closest statistical accent model to said statistical speech model;

assigning an accent score for the statistical speech model, the accent score corresponding to the model accent score of the closest statistical accent model;

automatically amending in the speech sample by replacement waveforms, sections determined as having known errors, whereby the amended speech sample replaces the speech sample to conform to the closest accent model; and

providing during the call conducted by the call responder an immediate visual feedback indicating the level of deviation of the agent from the required accent.

12. A method according to claim 11, wherein said determining is performed substantially in real-time.

13. A method according to claim 11, further comprising giving indication regarding segments of said speech sample which do not conform to the determined closest statistical accent model.

14. A method according to claim 13, wherein said indication is given substantially in real-time.

15. A method according to claim 11, further comprising notifying a speaker if the determined closest statistical accent changes during a conversation.

16. A method according to claim 11, wherein said determining further determines other characteristics of a speaker.

17. A method according to claim 11, wherein said one or more statistical accent models are updated based on speech signals from groups of substantially equally scoring users.

18. A method according to claim 11 wherein said accents are a pronunciation characteristic common to a group of people.

19. A method according to claim 11 wherein said accent is an acoustic speech feature that differentiates between individuals.

20. The method of claim 11, wherein the statistical accent models are created using at least one feature selected from the group consisting of:

phonology, acoustic phonetics, and phonotactics.

21. The method according to claim 11, further providing a supervisor of the call responder with quality information feedback related to the accent score of the call responder.

22. The method according to claim 11, further comprising asking a caller to repeat a specific sentence with pre-selected sounds that are known to differentiate between different accents.