Subject: Speech Processing

Topic: Speech Processing Applications

Difficulty: Low

(i) Tremendous advances have been observed in ASR systems in the past decades. The word recognition error rates has reduced by factor of 5.

(ii) Fast recognition algorithm have been developed into increase the recognition rate several time.

(iii) Speaker-independent speech recognition is made possible with high accuracy without extra hardware.

(iv) Continuous speech-recognition is also advanced in the following respects:

a) Speech modelling techniques have been improved to give higher speech recognition accuracy.

b) Search strategies have been developed to reduced the recognition time.

(v) These advances have occurred not only in laboratories, but speech recognition is strongly influencing the communication between human and machines.

(vi) Recent speech recognition uses probabilistic modelling methods, such as Hidden Markov Modelling (HMM).

(vii) These methods can also model the time variation of the spectral characteristics of speech signals. The model parameters can be found automatically from the training data. The probabilistic process has optimized the recognition accuracy.

(viii) The recognition error rates are found to reduce with the probabilistic modelling.

(ix) It is required to continue the research to find better models that are robust with respect to speaker variability, noise acoustic distortions.