Silvern (1981) On the use of dynamic time warping for word spotting and connected word recognition. [Bell Syst. Tech. J. 60, 3 (March 1981),303-325.]

getCITED

Home Search Add Content Reports Help

Edit Publication | Edit Contributors | Delete Publication | Edit References | Edit Citations

Add to Bookstack | Show Bookstack | Change Bookstack

On the use of dynamic time warping for word spotting and connected word recognition. [Bell Syst. Tech. J. 60, 3 (March 1981),303-325.]

Post a Comment

CONTRIBUTORS:

	Reviewer:	Silvern, Leonard C. (Systems Engineering Laboratories)
	Author:	MYERS, C. S. (MITCambridgeMass.) ;
	Co-investigator:	RABINER, L. R.;
	Co-investigator:	ROSENBERG, A. E.

JOURNAL:

Computing Reviews, 22(9), 398 - 398.

YEAR:

1981

PUB TYPE:

Book Review

SUBJECT(S):

Pattern Recognition

DISCIPLINE:

Computer Science

HTTP:

LANGUAGE:

English

PUB ID:

103-426-936 (Last edited on 2006/05/26 15:23:57 GMT-6)

ABSTRACT:
There is great interest today in methods of analyzing speech inputs to computer systems. This is in contrast to methods of synthesizing or creating voiced outputs. The authors focus on theoretical foundations for matching unknown words with known word patterns, so the outcome will be either recognition or non-recognition of a particular word or phrase.

The paper reviews basic principles of Dynamic Time Warping (DTW) as they apply to the recognition of discrete words and shows how DTW can be modified for both connected word recognition and word spotting. Matches normally seeking beginning and end points are now expected in the middle of a word and not necessarily near the end. In the past, the basic DTW algorithm considered beginning and ending frames, but the authors now emphasize beginning and ending regions to avoid problems of accurate segmentation of the utterance.

Experiments were conducted to compare fixed range and local minimum DTW algorithms, and also to study parameters of the local minimum algorithm. By varying several of the parameters, it was possible to achieve string recognition rates of nearly 90 percent. The paper contends that the local minimum DTW algorithm produces performance superior to the fixed range DTW algorithm.

While not a breakthrough in the speech analysis field, the paper does represent another well-designed brick in the foundation, and is worthy of consideration by communications engineers and computer programmers interested in chip and circuit design and programs for speech recognition.

L. C. Silvern, Sedona, Ariz

STATISTICS

Click on # to view
Citations		0
References		0
Comments		0
Quality		0/0.00
Interest		0/0.00
View(er)s		1/83

Quality
	N/A
High
	7
	6
	5
	4
	3
	2
	1
Low

Interest
	N/A
High
	7
	6
	5
	4
	3
	2
	1
Low

Prev | Next