Method for detecting emotions from speech using speaker identification

• US 7,373,301 B2
• Filed: 07/31/2002
• Issued: 05/13/2008
• Est. Priority Date: 08/02/2001
• Status: Active Grant

• Current Assignee: Sony Deutschland GmbH (Sony Group Corp.)
• Original Assignee: Sony Deutschland GmbH (Sony Group Corp.)
• Inventors: Kompe, Ralf, Kemp, Thomas, Tato, Raquel
• Primary Examiner(s): Vo; Huyen X.
• Application Number:
  US10/209,134
• Publication Number:
  US 20030028384A1
• Time in Patent Office: 2,113 Days
• Field of Search: 704/270, 704/275, 704/231, 704/246, 704/251, 704/273, 704/235, 704/245, 704/270.1, 713/100
• US Class Current: 704/275
• CPC Class Codes: G10L 17/26 Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices

Abstract

To reduce the error rate when classifying emotions from an acoustical speech input (SI) only, it is suggested to include a process of speaker identification to obtain certain speaker identification data (SID) on the basis of which the process of recognizing an emotional state is adapted and/or configured. In particular, speaker-specific feature extractors (FE) and/or emotion classifiers (EC) are selected based on said speaker identification data (SID).

Claims

1. A method for detecting emotions from speech input of at least one speaker,

identifying a given speech input so as to obtain speaker identification;

recognizing an emotional state of said speaker from said speech input by selecting a speaker-independent emotion recognizer when a current speaker has not been identified, and selecting a speaker-dependent emotion recognizer when a current speaker is identified; and

wherein the speaker-independent emotion recognizer includes emotion models trained in advance by a plurality of speakers, and wherein the speaker-dependent emotion recognizer includes emotion models trained in advance by the identified speaker.

2. The method according to claim 1,

wherein a predetermined set of speaker-specific emotion recognizers (SSER) is employed, and

each of said speaker-specific emotion recognizers (SSER) is assigned to a specific speaker.

3. The method according to claim 2, wherein in each of said speaker-specific emotion recognizers (SSER) at least one process of feature extraction or a feature extractor (FE) and one process of emotional classification or an emotion classificator (EC) are involved.

4. The method according to claim 2, wherein said process of recognizing an emotional state is adapted or configured by selecting a speaker-specific feature extractor (FE) or a speaker-specific emotion classificator (EC) based on said speaker identification.

5. The method according claim 1, wherein a speaker data base (SDB) of speakers already described or identified is employed in the process of speaker identification.

6. The method according to claim 1, wherein said process of speaker identification is carried out in a preprocessing step, on the entire speech input (SI) or a part thereof.

7. The method according to claim 1, wherein said processes of speaker identification and of adapting the emotion recognition are carried out essentially continuously or in a real-time or online manner, so as to increase the rate of speaker identification or so as to consider speaker changes.

8. The method according to claim 1, wherein said process of recognizing an emotional state is carried out on the entire speech input (SI) in a post-processing step, so as to increase the rate of emotion detection.

9. The method according to claim 1, wherein said process of recognizing an emotional state is carried out essentially continuously or in a real-time or online manner, so as to consider changes in the emotional state or to consider an emotional orbit or time-course.

10. The method according to claim 1 which is configured to consider a plurality of essentially simultaneous speakers so as to follow a time-course of the emotional states or the emotional orbits of the speakers during a conversation and a discussion.

11. A system for detecting emotions from a speech input, which is capable of performing and realizing a method for detecting emotions according to claim 1 and the steps thereof.