High function speech recognition systems generally require high quality audio input. They also require substantial computing power, RAM and disk storage, beyond the level that is practical in handsets for mobile networks. However, most speech recognition systems have a feature extraction front end which performs a significant level of data reduction, to a bit rate capable of being transmitted over most fixed and mobile networks. The project seeks to standardise this front end, allowing a wide range of different speech recognition systems and applications to provide full support to mobile and cordless handset users. A Mel Cepstrum algorithm is proposed for the feature extraction.
