Continuous Speech Recognition Using Hidden Markov Models

Continuous Speech Recognition Using Hidden Markov Models -...

Info iconThis preview shows pages 1–3. Sign up to view the full content.

View Full Document Right Arrow Icon
Continuous Speech Recognition Using Hidden Markov Models Joseph Picone Stochastic signal processing techniques have pro- foundly changed our perspective on speech processing. We have witnessed a progression from heuristic algo- rithms to detailed statistical approaches based on itera- tive analysis techniques. Markov modeling provides a mathematically rigorous approach to developing robust statistical signal models. Since the introduction of Markov models to speech processing in the middle 1970s. continuous speech recognition technology has come of age. Dramatic advances have been made in characterizing the temporal and spectral evolution of the speech signal. At the same time, our appreciation of the need to explain complex acoustic manifestations by integration of application constraints into low level signal processing has grown. In this paper, we review the use of Markov models in continuous speech recogni- tion. Markov models are presented as a generalization of its predecessor technology, Dynamic Programming. A unified view is offered in which both linguistic decoding and acoustic matching are integrated into a single opti- mal network search framework. 26 IEEE ASSP MAGAZINE JULY 1990 0740-7467/90/0700-0026$1.00 0 1990 IEEE .-.
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full Document Right Arrow Icon
Though automatic speech understanding by machine remains a distant goal in speech research, great strides have been made recently in the development of constrained, or application- specific, continuous speech recognition systems. Despite the fact that spoken language recognition still awaits more funda- mental breakthroughs in linguistics, we are witnessing the emergence of structural methods 111 that promise to be the foundation upon which future speech understanding systems will be built. At the core of this new generation of technology are powerful statistical signal processing approaches that inte- grate detailed statistical characterizations of the acoustic signal with probabilistic models of application constraints. In this review, we will restrict our scope to one particular class of statistical signal processing algorithms: first order Hidden Markov models (HMMs). Other variations 11-71 and generaliza- tions [8-91 hold great promise towards extending the frontier of speech recognition technology, and share similar foundations in statistical estimation theory. We present Hidden Markov model- ing as a generalization of its predecessor technology, Dynamic Programming (DP) [10,111. A unified view is offered in which both linguistic decoding and acoustic matching are integrated into a single optimal network search framework. Advances in Recognition Architectures The goal in continuous speech recognition is to provide a transcription of the incoming speech utterance, as depicted in Fig. 1. In this example, the speech signal has been decomposed into a sequence of phonetic units. This decomposition was the result of considering many possible explanations of the speech data given a model of all possible sentences that could be spo- ken, and choosing the best sequence based on some estimate of its likelihood. Information about the possible sentences that
Background image of page 2
Image of page 3
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

Page1 / 16

Continuous Speech Recognition Using Hidden Markov Models -...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online