Speech recognition

From A Cat's Wiki

Speech recognition means the process of converting a specific speech signal to a sequence of words. This is done by a computer program.

A related process is the speaker recognition, which tries to identify the person who is speaking.

[edit] Performance

To specify the performance of the speech recognition system you have to evaluate the accuracy and the speed. The accuracy can be measured with the word error rate. The real time factor measures the speed.

Commercially available dictation systems require a short period of training. They can have a very high accuracy (up to 99% accuracy) under optimal conditions. Optimal conditions mean that there is a quiet office available, the speaker has to speak clearly, a good microphone and a good sound card are needed. If the speaker has a strong accent, the recognition rate might be very low.

Dragon NaturallySpeaking 9.5 is currently (December 2007) probably the best commercially available speech recognition software.