1. recognition tests with added independent pink noise.
2. recognition tests with different S/N ratios.
3. recognition tests with a larger phoneme set.
4. recognition tests training HMM on both clean and noisy tokens.
5. recognition tests with different reference vector sizes.
6. optimization of the auditory model parameters.
In this report, we will focus on the first three problems. The 4th and 5th are the problems concerning the HMM phoneme recognition system itself. As they are topics whish are interesting from a practical viewpoint, we will leave them for the speech recognition people. The 6th is a fundamental problem. However, all experiments should be repeated when the auditory model is returned. Therefore, we decided not to do it at the moment.