mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-29 19:54:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update results.

Browse Source
This commit is contained in:
Fangjun Kuang 2021-12-27 15:31:35 +08:00
parent b5735ae16f
commit 8eea3414a4
4 changed files with 23 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
.github/workflows/run-pretrained-transducer-stateless.yml vendored
View File

@ -74,11 +74,11 @@ jobs:
mkdir tmp
cd tmp
git lfs install
git clone https://huggingface.co/csukuangfj/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22
git clone https://huggingface.co/csukuangfj/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27
cd ..
tree tmp
soxi tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/test_wavs/*.wav
ls -lh tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/test_wavs/*.wav
soxi tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/test_wavs/*.wav
ls -lh tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/test_wavs/*.wav
- name: Run greedy search decoding
shell: bash
@ -87,11 +87,11 @@ jobs:
cd egs/librispeech/ASR
./transducer_stateless/pretrained.py \
--method greedy_search \
--checkpoint ./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/exp/pretrained.pt \
--bpe-model ./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/data/lang_bpe_500/bpe.model \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/test_wavs/1089-134686-0001.wav \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/test_wavs/1221-135766-0001.wav \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/test_wavs/1221-135766-0002.wav
--checkpoint ./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/exp/pretrained.pt \
--bpe-model ./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/data/lang_bpe_500/bpe.model \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/test_wavs/1089-134686-0001.wav \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/test_wavs/1221-135766-0001.wav \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/test_wavs/1221-135766-0002.wav
- name: Run beam search decoding
shell: bash
@ -101,8 +101,8 @@ jobs:
./transducer_stateless/pretrained.py \
--method beam_search \
--beam-size 4 \
--checkpoint ./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/exp/pretrained.pt \
--bpe-model ./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/data/lang_bpe_500/bpe.model \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/test_wavs/1089-134686-0001.wav \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/test_wavs/1221-135766-0001.wav \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-22/test_wavs/1221-135766-0002.wav
--checkpoint ./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/exp/pretrained.pt \
--bpe-model ./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/data/lang_bpe_500/bpe.model \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/test_wavs/1089-134686-0001.wav \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/test_wavs/1221-135766-0001.wav \
./tmp/icefall-asr-librispeech-transducer-stateless-bpe-500-2021-12-27/test_wavs/1221-135766-0002.wav

2
README.md
View File

@ -84,7 +84,7 @@ The best WER using beam search with beam size 4 is:
| | test-clean | test-other |
|-----|------------|------------|
| WER | 2.92 | 7.37 |
| WER | 2.83 | 7.19 |
Note: No auxiliary losses are used in the training and no LMs are used
in the decoding.

18
egs/librispeech/ASR/RESULTS.md
View File

@ -4,7 +4,7 @@
#### Conformer encoder + embedding decoder
Using commit `fb6a57e9e01dd8aae2af2a6b4568daad8bc8ab32`.
Using commit `TODO`.
Conformer encoder + non-current decoder. The decoder
contains only an embedding layer and a Conv1d (with kernel size 2).
@ -13,12 +13,8 @@ The WERs are
| | test-clean | test-other | comment |
|---------------------------|------------|------------|------------------------------------------|
| greedy search | 2.99 | 7.52 | --epoch 20, --avg 10, --max-duration 100 |
| beam search (beam size 2) | 2.95 | 7.43 | |
| beam search (beam size 3) | 2.94 | 7.37 | |
| beam search (beam size 4) | 2.92 | 7.37 | |
| beam search (beam size 5) | 2.93 | 7.38 | |
| beam search (beam size 8) | 2.92 | 7.38 | |
| greedy search | 2.85 | 7.30 | --epoch 29, --avg 13, --max-duration 100 |
| beam search (beam size 4) | 2.83 | 7.19 | |
The training command for reproducing is given below:
@ -36,12 +32,12 @@ export CUDA_VISIBLE_DEVICES="0,1,2,3"
```
The tensorboard training log can be found at
<https://tensorboard.dev/experiment/PsJ3LgkEQfOmzedAlYfVeg/#scalars&_smoothingWeight=0>
<https://tensorboard.dev/experiment/Mjx7MeTgR3Oyr1yBCwjozw/>
The decoding command is:
```
epoch=20
avg=10
epoch=29
avg=13
## greedy search
./transducer_stateless/decode.py \
@ -64,7 +60,7 @@ avg=10
#### Conformer encoder + LSTM decoder
Using commit `TODO`.
Using commit `8187d6236c2926500da5ee854f758e621df803cc`.
Conformer encoder + LSTM decoder.

4
egs/librispeech/ASR/transducer_stateless/decode.py
View File

@ -70,14 +70,14 @@ def get_parser():
parser.add_argument(
"--epoch",
type=int,
default=20,
default=29,
help="It specifies the checkpoint to use for decoding."
"Note: Epoch counts from 0.",
)
parser.add_argument(
"--avg",
type=int,
default=10,
default=13,
help="Number of checkpoints to average. Automatically select "
"consecutive checkpoints before the checkpoint specified by "
"'--epoch'. ",