Experimental frameworks
Speaker-dependent speech recognition:
- the speech recognition system is trained for each speaker; the test audio-visual sequence belongs to a speaker already “known” by the speech recognition system.
- Note: the audio-visual test sequence is however different (i.e. different instance of time; maybe different conditions) than the one presented to the system in the training phase for the same speaker.
-
Speaker-independent speech recognition:
- the test audio-visual sequence is produced by an “unknown” speaker, who wasn’t presented to the speech recognition system in the training phase. This test is more difficult for the speech recognition system than the speaker-dependent test.
Department of Informatics
Aristotle University of Thessaloniki