mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-09-03 22:24:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update RESULTS.md

Browse Source
This commit is contained in:
zr_jin 2024-02-20 14:59:00 +08:00 committed by GitHub
parent 70e70edc8a
commit f79c5e15bc
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 35 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

71
egs/spgispeech/ASR/RESULTS.md
View File

@ -14,7 +14,7 @@ The WERs are:
|---------------------------|------------|------------|------------------------------------------| |---------------------------|------------|------------|------------------------------------------|
| greedy search | 2.08 | 2.14 | --epoch 30 --avg 10 | | greedy search | 2.08 | 2.14 | --epoch 30 --avg 10 |
| modified beam search | 2.05 | 2.09 | --epoch 30 --avg 10 --beam-size 4 | | modified beam search | 2.05 | 2.09 | --epoch 30 --avg 10 --beam-size 4 |
n| fast beam search | 2.07 | 2.17 | --epoch 30 --avg 10 --beam 20 --max-contexts 8 --max-states 64 | | fast beam search | 2.07 | 2.17 | --epoch 30 --avg 10 --beam 20 --max-contexts 8 --max-states 64 |
**NOTE:** SPGISpeech transcripts can be prepared in `ortho` or `norm` ways, which refer to whether the **NOTE:** SPGISpeech transcripts can be prepared in `ortho` or `norm` ways, which refer to whether the
transcripts are orthographic or normalized. These WERs correspond to the normalized transcription transcripts are orthographic or normalized. These WERs correspond to the normalized transcription
@ -39,31 +39,30 @@ The decoding command is:
``` ```
# greedy search # greedy search
python ./zipformer/decode.py \ python ./zipformer/decode.py \
--epoch $epoch \ --epoch $epoch \
--avg $avg \ --avg $avg \
--exp-dir ./zipformer/exp \ --exp-dir ./zipformer/exp \
--max-duration 1000 \ --max-duration 1000 \
--decoding-method modified_beam_search --decoding-method greedy_search
--decoding-method greedy_search
# modified beam search # modified beam search
python ./zipformer/decode.py \ python ./zipformer/decode.py \
--epoch $epoch \ --epoch $epoch \
--avg $avg \ --avg $avg \
--exp-dir ./zipformer/exp \ --exp-dir ./zipformer/exp \
--max-duration 1000 \ --max-duration 1000 \
--decoding-method modified_beam_search --decoding-method modified_beam_search
# fast beam search # fast beam search
python ./zipformer/decode.py \ python ./zipformer/decode.py \
--epoch $epoch \ --epoch $epoch \
--avg $avg \ --avg $avg \
--exp-dir ./zipformer/exp \ --exp-dir ./zipformer/exp \
--max-duration 1000 \ --max-duration 1000 \
--decoding-method fast_beam_search --decoding-method fast_beam_search
--beam 4 \ --beam 4 \
--max-contexts 4 \ --max-contexts 4 \
--max-states 8 --max-states 8
``` ```
### SPGISpeech BPE training results (Pruned Transducer) ### SPGISpeech BPE training results (Pruned Transducer)
@ -109,28 +108,28 @@ The decoding command is:
``` ```
# greedy search # greedy search
./pruned_transducer_stateless2/decode.py \ ./pruned_transducer_stateless2/decode.py \
--iter 696000 --avg 10 \ --iter 696000 --avg 10 \
--exp-dir ./pruned_transducer_stateless2/exp \ --exp-dir ./pruned_transducer_stateless2/exp \
--max-duration 100 \ --max-duration 100 \
--decoding-method greedy_search --decoding-method greedy_search
# modified beam search # modified beam search
./pruned_transducer_stateless2/decode.py \ ./pruned_transducer_stateless2/decode.py \
--iter 696000 --avg 10 \ --iter 696000 --avg 10 \
--exp-dir ./pruned_transducer_stateless2/exp \ --exp-dir ./pruned_transducer_stateless2/exp \
--max-duration 100 \ --max-duration 100 \
--decoding-method modified_beam_search \ --decoding-method modified_beam_search \
--beam-size 4 --beam-size 4
# fast beam search # fast beam search
./pruned_transducer_stateless2/decode.py \ ./pruned_transducer_stateless2/decode.py \
--iter 696000 --avg 10 \ --iter 696000 --avg 10 \
--exp-dir ./pruned_transducer_stateless2/exp \ --exp-dir ./pruned_transducer_stateless2/exp \
--max-duration 1500 \ --max-duration 1500 \
--decoding-method fast_beam_search \ --decoding-method fast_beam_search \
--beam 4 \ --beam 4 \
--max-contexts 4 \ --max-contexts 4 \
--max-states 8 --max-states 8
``` ```
Pretrained model is available at <https://huggingface.co/desh2608/icefall-asr-spgispeech-pruned-transducer-stateless2> Pretrained model is available at <https://huggingface.co/desh2608/icefall-asr-spgispeech-pruned-transducer-stateless2>