Abstract:

In order to minimize the recognition errors caused by inaccurate model estimations from those toned triphones with limited training samples, we proposed to initialize toned triphones using their own toneless triphone model parameters. Besides, works concerning stopping criteria of decision tree state tying as well as mixture component adaptation are also explored to obtain better performance as well as reduce model scale. Experiments results have shown that, on the 863 corpus, along with all this improvements our system achieves certain increase of phone recognition rate, with much more trainable model scale as well.

Key words: acoustic model, speech recognition, triphone, tone