In this paper we present a discussion on the use of a discrete wavelet transform in neural network speech recognition problems, and this has allowed us to use relatively small neural network architectures for speech recognition tasks. We also demonstrate the ability of a neural network to recognize compressed speech using the wavelet approach. An additional topic of this paper is the speech compression alogrithm that was developed for the particular application of maintaining the defining characteristics of a persons voice, including pitch and prosody. Speeding up spoken words without losing the speaker's voice characteristics has applications in learning, where some researchers have theorized that if sentences are spoken faster some students will learn better. Examples demonstrate the ability of the speech recognizer to identify the compressed-time speech.
展开▼
