mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-09-19 05:54:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

add ngram-lm-scale=0.35 results

Browse Source
This commit is contained in:
luomingshuang 2022-07-26 14:56:03 +08:00
parent 749dfcdb7d
commit 93f4eb1164
2 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
egs/wenetspeech/ASR/RESULTS.md
View File

@ -15,7 +15,7 @@ When training with the L subset, the WERs are
| fast beam search (1best) | 7.94 | 8.74 | 13.80 | --epoch 10, --avg 2, --max-duration 1500 |
| fast beam search (nbest) | 9.82 | 10.98 | 16.37 | --epoch 10, --avg 2, --max-duration 600 |
| fast beam search (nbest oracle) | 6.88 | 7.18 | 11.77 | --epoch 10, --avg 2, --max-duration 600 |
| fast beam search (nbest LG) | 14.94 | 16.14 | 22.93 | --epoch 10, --avg 2, --max-duration 600 |
| fast beam search (nbest LG, ngram_lm_scale=0.35) | 8.83 | 9.88 | 15.47 | --epoch 10, --avg 2, --max-duration 600 |
The training command for reproducing is given below:
@ -110,6 +110,7 @@ avg=2
--lang-dir data/lang_char \
--max-duration 600 \
--decoding-method fast_beam_search_nbest_LG \
--ngram-lm-scale 0.35 \
--beam 20.0 \
--max-contexts 8 \
--max-states 64

4
egs/wenetspeech/ASR/pruned_transducer_stateless2/decode.py
View File

@ -225,7 +225,7 @@ def get_parser():
parser.add_argument(
"--ngram-lm-scale",
type=float,
default=0.01,
default=0.07,
help="""
Used only when --decoding_method is fast_beam_search_nbest_LG.
It specifies the scale for n-gram LM scores.
@ -590,6 +590,8 @@ def main():
params.suffix += f"-beam-{params.beam}"
params.suffix += f"-max-contexts-{params.max_contexts}"
params.suffix += f"-max-states-{params.max_states}"
if params.decoding_method == "fast_beam_search_nbest_LG":
params.suffix += f"-ngram-lm-scale-{params.ngram_lm_scale}"
elif "beam_search" in params.decoding_method:
params.suffix += f"-beam-{params.beam_size}"
else: