Samy Bengio
We describe a new method for phoneme sequence recognition given a speech utterance. In contrast to HMM-based approaches, our method uses a kernel-based discriminative training procedure in which the learning process is tailored to the goal of minimizing th ...
IDIAP2006