Browse Prior Art Database

Speech Information Retrieval System

IP.com Disclosure Number: IPCOM000114655D
Original Publication Date: 1995-Jan-01
Included in the Prior Art Database: 2005-Mar-29
Document File: 2 page(s) / 34K

Publishing Venue

IBM

Related People

Sugawara, K: AUTHOR

Abstract

Disclosed is a system for storing and retrieving speech data. It consists of recording, play-back, and speech-data-searching components. Recorded data are analyzed and vector quantized to create a sequence of phonetic labels for short time (10ms) segments. To retrieve recorded speech data, the system processes key input as recorded data and converts these data into a sequence of labels. Searching is done by means of the continuous matching method, which allows temporal and spectral fluctuations. The search result is a list of candidate segments with similarity scores. It can be displayed to the user as a waveform or played back.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 100% of the total text.

Speech Information Retrieval System

      Disclosed is a system for storing and retrieving speech data.
It consists of recording, play-back, and speech-data-searching
components.  Recorded data are analyzed and vector quantized to
create a sequence of  phonetic labels for short time (10ms) segments.
To retrieve recorded speech data, the system processes key input as
recorded data and converts these data into a sequence of labels.
Searching is done by means of the continuous matching method, which
allows temporal and spectral fluctuations.  The search result is a
list of candidate segments with similarity scores.  It can be
displayed to the user as a waveform or played back.

      The Figure shows a diagram of the system.  Input data are
stored in (2) as waveform data.  They are also analyzed by using a
phonetic label database (8), and stored in (5) as a sequence of
labels.  Key speech input is also analyzed and converted into a label
sequence (10).  The continuous matching engine (9) searches through
the stored label data (5) and returns a list of matching candidates
in (11).