I want to mix two LM, one in word and another sub-lexical item (e.g. a syllable).
It ultimately depends on what you want to do...
SRILM has a tool for model interpolation. But this usually implies having a common symbol set shared by the models.
See http://www.speech.sri.com/projects/srilm/manpages/ngram.1.html
The following paper describes the use of hybrid models (mixing different classes of symbol in the same model), as applied to OOV detection:
L. Qin and A. Rudnicky OOV Word Detection using Hybrid Models with Mixed Types of Fragments. Proceedings of Interspeech, 2012, Portland, OR US.
Finally, you can decode the speech multiple times, each time using a different model. Of course you then have to figure out alignment and possible fusion, if that's relevant.
It really does depend on what you're after. Perhaps you can be more specific about your goal.
- Badges
- Science topic
- Similar topics
- Psychology