User's Manual

DATASHEET FC6000 Confidential Information 10/53
2.5 Speaker independent Voice recognition
2.5.1 Voice Recognition principles
2.5.1.1 Description
VoCon 3200 V3.3 is NUANCE training-less speaker independent speech recognition
algorithm.
• Phonemes recognition: words are recognized without previous training
• Words models for a better precision, especially for digits recognition
• Continuous voice recognition: no need for blanks between words
• New words learning (Voice tags), speaker dependent speech recognition (100 Voice
tags, 2kbytes by Voice tag)
• Noise robustness and accuracy in an automotive environment: engine, click-button
etc…
• Highly accurate recognition
• VoCon Music Pre-Processor. This feature allows the user to select music to play by
voice commands.
2.5.1.2 Operation
During a voice recognition process, "Feature Extraction" module decomposes the signal. The
module "Search" looks for the equivalent text using the modules "G2P" and "Grammar to
compile". These two modules are using the libraries "Acoustic Model", "Lexicon" and
"Grammar".
Module G2P ensures equivalence between the graphemes and the phonemes.
For each language is associated an acoustic model ("Acoustic Model"), a grammar and a
lexicon ("Grammar" and "Lexicon").
System feedback is realized by a screen display and/or sounds (synthesized voice, chime,
pre-recorded voice prompts…).
Operation is ended by a final action (phone number dialing, radio station tuning…).