search

k2-fsa / snowfall

Moved to https://github.com/k2-fsa/icefall
Apache License 2.0
143 stars 42 forks source link

Add timit recipe #247

Open luomingshuang opened 2 years ago

luomingshuang commented 2 years ago

I set subsampling_factor=2.

TIMIT CTC_Train

Testing results based on different training epochs:

epoch=20
2021-09-03 10:54:10,903 INFO [ctc_decode.py:188] %PER 30.34% [2225 / 7333, 293 ins, 441 del, 1491 sub ]

epoch=30
2021-09-03 10:59:10,147 INFO [ctc_decode.py:188] %PER 29.77% [2183 / 7333, 221 ins, 473 del, 1489 sub ]

epoch=35
2021-09-03 11:11:00,885 INFO [ctc_decode.py:188] %PER 28.94% [2122 / 7333, 266 ins, 397 del, 1459 sub ]

epoch=40
2021-09-03 11:12:39,029 INFO [ctc_decode.py:188] %PER 29.52% [2165 / 7333, 304 ins, 348 del, 1513 sub ]
csukuangfj commented 2 years ago

https://github.com/speechbrain/speechbrain/tree/develop/recipes/TIMIT/ASR/CTC

The PER in speechbrain is about 14.78. Seems there is a lot of room for improvement

luomingshuang commented 2 years ago

Yes, I am having a experiment with mmi, and I will have a look at their codes.

https://github.com/speechbrain/speechbrain/tree/develop/recipes/TIMIT/ASR/CTC

The PER in speechbrain is about 14.78. Seems there is a lot of room for improvement