An Integration of Hidden Markov Model and Neural Network for Phoneme Recognition

Speech recognition has been a popular research topic in the past 20 years. Different approaches have been attempted by other research and they exhibit some advantages over another. Hidden Markov Model (HMM) is one common approach iliat has been used in many researches for the past ten years. HMM is a stochastic process which provides an efficient means of modelling the sequential structure such as speech. However, in the past five years HMM approach has been evaluated and its discrimination problem has been another popular topic. In our research, we based on our interpretation of this problem and developed a new model for phoneme recognition. Our model transforms the problem of recognizing dynamic sequential patterns using HMM into a static pattern recognition problem using an integrated HMM-Neural Network approach. We carried out our experiments using the TIMIT multiple speakers speech database. We compared our approach with the HMM approach using 600 speech samples in six phoneme classes. For a training/testing data ratio of 300/300, the integrated approach obtained an increase of 1.3% in recognition rate over the HMM. When the trainin^testing data ratio became 450/150, the integrated approach obtained an increase of 4.7% in recognition rate over the HMM. Based on these results, we conclude that a neural network is justified to partially solve the HMM discrimination problem. Acknowledgements It has been an enjoyable experience to be involved in the speech recognition project supervised by Dr. Lai Wan Chan. She is always patient to explain and discuss the difficulties and problems that we have encountered in this project. Without her valuable advice and insight, this thesis would not have been completed. Table of

• Publication year

  2013

• Venue

  Unknown venue

• Publication date

  Unknown publication date

• Fields of study

  Not labeled

• Identifiers

  No identifiers available.

• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Automatic Speech Recognition: The Development of the SPHINX System

  2013cited by this paper
• Alpha-nets: A recurrent 'neural' network architecture with a hidden Markov model interpretation

  1990cited by this paper
• Speaker-independent phoneme recognition on TIMIT database using integrated time-delay neural networks (TDNNs)

  1990cited by this paper
• Continuous speech recognition using multilayer perceptrons with hidden Markov models

  1990cited by this paper
• Links Between Markov Models and Multilayer Perceptrons

  1990cited by this paper
• Speaker-independent phone recognition using hidden Markov models

  1989cited by this paper
• Phoneme recognition using time-delay neural networks

  1989influential reference
• Review of Neural Networks for Speech Recognition

  1989cited by this paper
• An introduction to hidden Markov models

  1986cited by this paper
• Recognition of isolated digits using hidden Markov models with continuous mixture densities

  1985cited by this paper
• The use of speech knowledge in automatic speech recognition

  1985influential reference
• An introduction to the application of the theory of probabilistic functions of a Markov process to automatic speech recognition

  1983influential reference
• On the application of vector quantization and hidden Markov models to speaker-independent, isolated word recognition

  1983cited by this paper
• Distortion performance of vector quantization for LPC voice coding

  1982cited by this paper
• An Algorithm for Vector Quantizer Design

  1980cited by this paper
• Evaluation of a word recognition system using syntax analysis

  1978cited by this paper
• Speech recognition by machine: A review

  1976influential reference
• Minimum Prediction Residual Principle Applied to Speech Recognition

  1975cited by this paper

• No citing papers are available for this paper.