mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-09-11 18:14:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update RESULTS.md

Browse Source
This commit is contained in:
uni-manjunath-ke 2023-05-31 15:18:54 +05:30 committed by GitHub
parent 01d8c92765
commit be47220dcb
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 33 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
egs/librispeech/ASR/RESULTS.md
View File

@ -6,38 +6,6 @@ See <https://github.com/k2-fsa/icefall/pull/1058> for more details.
[zipformer](./zipformer) [zipformer](./zipformer)
##### zipformer ctc streaming
| decoding method | test-clean | test-other | comment |
|----------------------|------------|------------|--------------------|
| greedy_search | 4.07 | 10.51 | --epoch 30 --avg 15|
| greedy_search | 4.0 | 10.3 | --epoch 30 --avg 9 |
The training command is:
```bash
export CUDA_VISIBLE_DEVICES="0,1"
python ./zipformer_ctc_streaming/train.py \
--world-size 2 \
--num-epochs 30 \
--start-epoch 1 \
--exp-dir ./zipformer_ctc_streaming/exp \
--max-duration 100
```
The decoding command is:
```bash
export CUDA_VISIBLE_DEVICES="0"
./zipformer_ctc_streaming/decode.py \
--epoch 30 \
--avg 15 \
--exp-dir ./zipformer_ctc_streaming/exp \
--max-duration 300 \
--decode-chunk-len 32 \
--method ctc-decoding \
--lm-dir data/lm \
--lang-dir data/lang_bpe_500
```
#### Non-streaming #### Non-streaming
##### normal-scaled model, number of model parameters: 65549011, i.e., 65.55 M ##### normal-scaled model, number of model parameters: 65549011, i.e., 65.55 M
@ -357,6 +325,39 @@ The decoding commands are:
--max-states 64 --max-states 64
``` ```
##### zipformer ctc streaming
| decoding method | test-clean | test-other | comment |
|----------------------|------------|------------|--------------------|
| greedy_search | 4.07 | 10.51 | --epoch 30 --avg 15|
| greedy_search | 4.0 | 10.3 | --epoch 30 --avg 9 |
The training command is:
```bash
export CUDA_VISIBLE_DEVICES="0,1"
python ./zipformer_ctc_streaming/train.py \
--world-size 2 \
--num-epochs 30 \
--start-epoch 1 \
--exp-dir ./zipformer_ctc_streaming/exp \
--max-duration 100
```
The decoding command is:
```bash
export CUDA_VISIBLE_DEVICES="0"
./zipformer_ctc_streaming/decode.py \
--epoch 30 \
--avg 15 \
--exp-dir ./zipformer_ctc_streaming/exp \
--max-duration 300 \
--decode-chunk-len 32 \
--method ctc-decoding \
--lm-dir data/lm \
--lang-dir data/lang_bpe_500
```
### pruned_transducer_stateless7 (zipformer + multidataset(LibriSpeech + GigaSpeech + CommonVoice 13.0)) ### pruned_transducer_stateless7 (zipformer + multidataset(LibriSpeech + GigaSpeech + CommonVoice 13.0))
See <https://github.com/k2-fsa/icefall/pull/1010> for more details. See <https://github.com/k2-fsa/icefall/pull/1010> for more details.