This projects is a experimental project, we implemented deepspeech2 on aishell dataset.
We get around 20% word err rate,and about 5% phoneme err rate:
Here are some recognition samples:
target:剃光头吧
predict:提供头伴
target:感叹时光匆匆
predict:但看时光丰松
target:我没家没小孩
predict:我们家没小孩
target:再度备受关注
predict:再度备手关注
target:十多人受伤
predict:石多人受伤
target:成为了幸运儿
predict:成为了幸玥儿
target:而近年的蓝洁瑛
predict:而今年的短节影
target:整个人焕然一新
predict:整个人换着一些
target:大家神情放松
predict:大家申请放松
target:获观众的心
predict:吕获观众的心