System and method for automated multimedia content indexing and retrieval

• US 6,714,909 B1
• Filed: 11/21/2000
• Issued: 03/30/2004
• Est. Priority Date: 08/13/1998
• Status: Expired due to Fees

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: Gibbon, David Crawford, Shahraray, Behzad, Liu, Zhu, Huang, Qian, Rosenberg, Aaron Edward
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): NOLAN, DANIEL A
• Application Number:
  US09/716,278
• Time in Patent Office: 1,225 Days
• Field of Search: 704/246, 704/260, 704/251, 704/270.1, 382/309, 345/719, 345/749, 725/134, 707/6, 707/3, 707/102
• US Class Current: 704/246
• CPC Class Codes: G06F 16/739 in form of a video summary, e.g. the video summary being a video sequence, a composite still image or having synthesized frames
• CPC Class Codes: G06F 16/7834 using audio features
• CPC Class Codes: G06F 16/7844 using original textual content or text extracted from visual content or transcript of audio data
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99943 Generating database or data structure, e.g. via user interface

Abstract

The invention provides a system and method for automatically indexing and retrieving multimedia content. The method may include separating a multimedia data stream into audio, visual and text components, segmenting the audio, visual and text components based on semantic differences, identifying at least one target speaker using the audio and visual components, identifying a topic of the multimedia event using the segmented text and topic category models, generating a summary of the multimedia event based on the audio, visual and text components, the identified topic and the identified target speaker, and generating a multimedia description of the multimedia event based on the identified target speaker, the identified topic, and the generated summary.

Claims

1. A method for automatically indexing and retrieving a multimedia event, comprising:

separating a multimedia data stream into audio, visual and text components;

segmenting the audio, visual and text components of the multimedia data stream based on semantic differences;

identifying at least one target speaker using the audio and visual components;

identifying semantic boundaries of text for at least one of the identified target speakers to generate semantically coherent text blocks;

generating a summary of multimedia content based on the audio, visual and text components, the semantically coherent text blocks and the identified target speaker;

deriving a topic for each of the semantically coherent text blocks based on a set of topic category models;

generating a multimedia description of the multimedia event based on the identified target speaker, the semantically coherent text blocks, the identified topic, and the generated summary; and

extracting audio features from the audio component of the multimedia data stream, the audio features being at least one of frame-level and clip level features,

wherein the frame level features in three subbands are at least one of volume, zero crossing rate, pitch period, frequency centroid, frequency bandwidth, and energy ratios.

2. A method for automatically indexing and retrieving a multimedia event, comprising:

separating a multimedia data stream into audio, visual and text components;

segmenting the audio, visual and text components of the multimedia data stream based on semantic differences;

identifying at least one target speaker using the audio and visual components;

identifying semantic boundaries of text for at least one of the identified target speakers to generate semantically coherent text blocks;

generating a summary of multimedia content based on the audio, visual and text components, the semantically coherent text blocks and the identified target speaker;

deriving a topic for each of the semantically coherent text blocks based on a set of topic category models;

generating a multimedia description of the multimedia event based on the identified target speaker, the semantically coherent text blocks, the identified topic, and the generated summary; and

extracting audio features from the audio component of the multimedia data stream, the audio features being at least one of frame-level and clip level features,

wherein clip level features are classified as at least one of time domain features and frequency domain features.

3. The method of claim 2, wherein the time domain features are at least one of non-silence ratio, volume standard deviation, standard deviation of zero crossing rate, volume dynamic range, volume undulation, and 4Hz modulation energy.

4. The method of claim 2, wherein the frequency domain features in three subbands are at least one of standard deviation of the pitch period, smooth pitch ratio, non-pitch ratio, frequency centroid, frequency bandwidth, and energy ratios.

5. A system that automatically indexes and retrieves a multimedia event, comprising:

a multimedia data stream separation unit that separates a multimedia data stream into audio, visual and text components;

a data stream component segmentation unit that segments the audio, visual and text components of the multimedia data stream based on semantic differences;

a target speaker detection unit that identifies at least one target speaker using the audio and visual components;

a content segmentation unit that identifies semantic boundaries of text, for at least one of the identified target speakers, to generate semantically coherent text blocks;

a summary generator that generates a summary of multimedia content based on the audio, visual and text components, the semantically coherent text blocks and the identified target speaker;

a topic categorization unit that derives a topic for each of the semantically coherent text blocks based on a set of topic category models;

a multimedia description generator that generates a multimedia description of the multimedia event based on the identified target speaker, the semantically coherent text blocks, the identified topic, and the generated summary; and

a feature extraction unit that extracts audio features from the audio component of the multimedia data stream, the audio features being at least one of frame-level and clip level features,

wherein the frame level features in three subbands are at least one of volume, zero crossing rate, pitch period, frequency centroid, frequency bandwidth and energy ratios.

6. The system of claim 5, wherein clip level features are classified as at least one of time domain features and frequency domain features.

7. The system of claim 6, herein the time domain features are at least one of non-silence ratio, volume standard deviation, standard deviation of zero crossing rate, volume dynamic range, volume undulation, and 4 Hz modulation energy.

8. The system of claim 6, wherein the frequency domain features in three subbands are at least one of standard deviation of the pitch period, smooth pitch ratio, non-pitch ratio, frequency centroid, frequency bandwidth, and energy ratios.