In the medium-term we plan to continue integrating into HTK3 a number
of the features that have been used at CUED in research systems. In
the longer term more advanced facilities developed at CUED will be
integrated in the released code base. This will include Cluster
Adaptive Training (CAT) and noise robustness approaches such as Vector
Taylor Series compensation, Joint Uncertainty Decoding.
We also hope to provide more speech recognition infrastructure in a
form that can be used with HTK. This will include for example word
lattices for standard speech recognition tasks.
Our medium term release plans are as follows:
- (End of 2009): release a recipe for an English large vocabulary
speech recognition task, including discriminative training.
If you'd like to provide help with any of the above tasks, please
contact us either by posting to the HTK-developers
mailing list or emailing us privately at email@example.com
Comments and suggestions to