Real-time implementation and evaluation of acoustic processing techniques for automatic speech recognition

Abstract The implementation and evaluation of an acoustic processor for automatic speech recognition using the K-means iteration vector quantization algorithm are described. To analyse the features obtained the ·chainmap’ procedure is used. The acoustic processor is implemented using a TMS320C25 DSP which interfaces directly to a transputer array. Requirements considered include the accuracy to represent the speech signal, real-time processing capacity and flexibility. Speech samples using five independent speakers are used to assess the acoustic features extracted by the processor. The system has been further developed to allow satisfactory processing using 12 speakers.