In this talk, I will discuss two speech recognition projects. The first project involves the development of a speech signal representation that explicitly targets the linguistic information in the speech signal. This representation consists of acoustic parameters (APs) which are exact measures performed on the speech signal or its time-frequency representation to provide evidence for the acoustic correlates of phonetic features. Our philosophy in developing the APs is that they must be based on relative measures to minimize speaker differences. We call this speech signal representation a knowledge-based front end (KBFE) since one of its important applications is for speech recognition. The second project is the development of an event-based speech recognition system (EBS). In this system, landmarks are extracted from the speech signal by the APs associated with the manner phonetic features. The landmarks are then used to extract information related to the place-of-articulation phonetic features and the source phonetic features. The net result of EBS will be an automatic recognition system that achieves high performance for limited vocabulary tasks without the need for training. This research is of theoretical and practical interest. It is of theoretical interest because EBS serves as a model of speech production. In addition, the development of EBS serves as a tool for studying speech variability. It is of practical interest because EBS has many applications beyond speech recognition, including (1) serving as a front end for various speech enhancement systems, (2) serving as an aid in speech analysis systems, (3) performing automatic time alignment of the speech waveform with phonetic labels, and (4) possibly serving as part of a speech training aid for speech pathologists and their clients.
For the colloquium series schedule, see the UMD Computational Linguistics Colloquium Series web page at http://umiacs.umd.edu/~resnik/cl_colloquium/. If you are interested in meeting with the speaker, please contact Philip Resnik (resnik@umiacs.umd.edu).