Time domain segmentation of speech signals

With the inclusion of computers in an increasing number of everyday activities, much effort is being made to improve communications between man and machine. One area of intense research is the computer recognition of human speech. A basic component of almost all speech recognition schemes is the capability to distinguish silence and noise from speech segments and voiced from unvoiced segments. This thesis examines the segmentation of isolated speech into voiced, unvoiced, and silence segments. The goal of the segmentation is to detect true boundaries rather than categorize fixed segments of time. Another goal is to produce a scheme that is applicable to real-time speech recognition. As such, the use of training data and prior knowledge of the input is avoided. To reduce computational overhead, frequency domain parameters are not used to arrive at initial segmentation decisions. It is recognized that some type of frequency domain processing, such as linear prediction, may be necessary in phonetic recognition. Phonetic recognition is outside the scope of this thesis.

Name thesis:degree_name

Master of Science

Level thesis:degree_level

Masters

Discipline thesis:degree_discipline

Computer Engineering

Year dc:date.issued

1997

Author dc:creator

• Huyck, Patrick John

Advisor dc:contributor.advisor

• Upda, Satish S.

Language dc:language.iso

en

DOI

https://doi.org/10.31274/td-20250603-322

OAI identifier oai:identifier

oai:dr.lib.iastate.edu:20.500.12876/Nveo5x5z