Central and Eastern European Survey
Research
Speech Recognition & Synthesis Laboratory (SRSL)
Institute of Engineering Cybernetics (IEC) National Ac.of Sc. of Belarus
Research Activities pursued at the organisation
The main scientific research
directions of the SRSL are the following:
- Spectrum and signal based features extraction;
- Formant analysis and synthesis;
- Microwaves speech signal synthesis;
- Rule-based text-to-speech synthesis;
- Expert knowledge-based recognition of visual spectrograms;
- Phonemic segmentation and labelling rules;
- Speech-to-text recognition;
- Discrete and connected words recognition;
- Word spotting;
- Computer telephony applications;
- Computer-based systems for the blind.
As a result of previous research projects the following soft & hardware
systems were created:
- a multi-lingual model of speech synthesis from text
based on the formants speech signal description.
Languages: Russian, English, German and French;
- a multilingual model of speech synthesis from text
based on the microwaves(MW) speech signal description.
Languages: Russian, Belarussian, Ukrainian, Polish and Czech;
- the PHONEMAPHONE - the first PC-based Russian text-to-speech
synthesizer providing high intelligibility and sufficient naturalness
of synthetic speech. It is utilized in the several commercial products:
telephone services, workstations for the blind, Russian oral speech
training systems, transport and some others applications;
- a knowledge-based phonemic interpretation model of the spoken
Russian words.
- a model of discret words recognition. It provides robust
real-time PC-based speaker-independent words (about 100) recognition and
is designed for computer telephony applications;
- a model of noise-resistant connected words recognition with
problem-oriented vocabulary designed for aircraft applications;
Current research projects:
- creation of a high quality text-to-speech synthesis model based
on allophonic wave-vorms speech signal representation and on syntax-
morphological analysis of reading text;
- creation of a DTW & HMM-based robust speech recognizer
providing speaker-and channel independent words recognition
for computer telephony applications;
|