No abstract

This paper presents a method for, a n d performance of, p h o n e m e s e g m e n t a t i o n b y a n e x p e r t s y s t e m u t i l i z i n g spectrogram reading strategy a n d knowledge. T h e e x p e r t system detects phonemes i n a spectrogram a n d determines their b o u n d a r i e s as well as their c o a r s e categories. T o simulate a human expert spectrogram reading process, the system performs assumption-based inference with certainty factors, a n d top-down acoustic f e a t u r e extraction u n d e r phonetic c o n t e x t hypotheses. T h e system, i n t o w h i c h J a p a n e s e c o n s o n a n t s e g m e n t a t i o n k n o w l e d g e is incorporated, is able to detect ahout 90% of the phonemes correctly. I n particular, t h e phoneme boundaries detected b y the system a r e as accurate as those detected b y h u m a n experts. T h e result is that the phonemes obtained b y t h e expert system can h e identified using a stochastic phoneme recognition method.

No abstract

No abstract

The Voice Activity Detection (VAD) problem is placed into a decision theoretic framework, and the Gaussian VAD model of Sohn et al. is then shown to fit well with the framework. It is argued that the Gaussian model can be made more robust to correlation and expected spectral shapes of speech and noise by using a differential spectral representation. Such a model is formulated theoretically. The differential spectral VAD is then shown by experiment to be consistently superior to the basic Gaussian VAD in a speech recognition setting, especially for noisy environments.