Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

add fleurs conformer+sc-ctc results #4746

Merged
merged 5 commits into from
Nov 4, 2022
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
36 changes: 36 additions & 0 deletions egs2/fleurs/asr1/README.md
Original file line number Diff line number Diff line change
@@ -1,4 +1,40 @@
<!-- Generated by scripts/utils/show_asr_result.sh -->
# Multilingual ASR - SSL + Conformer + Self-condition [XLS-R, Conformer, utt_mvn, 6500 BPE](conf/train_asr_conformer_scctc.yaml)

## Environments
- date: `Sat Oct 22 14:55:21 EDT 2022`
- python version: `3.8.6 (default, Dec 17 2020, 16:57:01) [GCC 10.2.0]`
- espnet version: `espnet 202207`
- pytorch version: `pytorch 1.8.1+cu102`
- Git hash: `e534106b837ff6cdd29977a52983c022ff1afb0f`
- Commit date: `Sun Sep 11 22:31:23 2022 -0400`
- Pretrained Model: https://huggingface.co/espnet/wanchichen_fleurs_asr_conformer_scctc

## asr_train_asr_conformer_scctc_raw_all_bpe6500_sp

### Language Identification
|dataset|Accuracy|
|---|---|
|decode_asr_lm_lm_train_lm_all_bpe6500_valid.loss.ave_asr_model_valid.acc.ave_3best/test_all|0.9541 (74237/77809)|

### WER

|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
|---|---|---|---|---|---|---|---|---|
|decode_asr_lm_lm_train_lm_all_bpe6500_valid.loss.ave_asr_model_valid.acc.ave_3best/test_all|77809|1592160|70.5|26.1|3.4|3.4|32.9|97.0|

### CER

|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
|---|---|---|---|---|---|---|---|---|
|decode_asr_lm_lm_train_lm_all_bpe6500_valid.loss.ave_asr_model_valid.acc.ave_3best/test_all|77809|10235271|92.2|4.7|3.1|2.6|10.4|97.0|

### TER

|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
|---|---|---|---|---|---|---|---|---|
|decode_asr_lm_lm_train_lm_all_bpe6500_valid.loss.ave_asr_model_valid.acc.ave_3best/test_all|77809|9622352|91.3|5.6|3.1|2.7|11.4|97.0|

# Multilingual ASR - Self-supervised learning features [HuBERT_large_ll60k, Transformer, utt_mvn, 6500 BPE](conf/train_asr_hubert_large_ll60k_transformer.yaml)

## Environments
Expand Down
2 changes: 1 addition & 1 deletion egs2/fleurs/asr1/conf/train_asr.yaml
89 changes: 89 additions & 0 deletions egs2/fleurs/asr1/conf/tuning/train_asr_conformer_scctc.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,89 @@
batch_type: numel
batch_bins: 12000000
accum_grad: 4
max_epoch: 10
patience: none
# The initialization method for model parameters
init: xavier_uniform
best_model_criterion:
- - valid
- acc
- max
keep_nbest_models: 3

encoder: conformer
encoder_conf:
output_size: 512
attention_heads: 8
linear_units: 2048
num_blocks: 12
dropout_rate: 0.1
positional_dropout_rate: 0.1
attention_dropout_rate: 0.1
input_layer: conv2d2
normalize_before: true
macaron_style: true
pos_enc_layer_type: "rel_pos"
selfattention_layer_type: "rel_selfattn"
activation_type: "swish"
use_cnn_module: true
cnn_module_kernel: 31
interctc_layer_idx: [3, 6, 9,]
interctc_use_conditioning: true

decoder: transformer
decoder_conf:
attention_heads: 8
linear_units: 2048
num_blocks: 6
dropout_rate: 0.1
positional_dropout_rate: 0.1
self_attention_dropout_rate: 0.1
src_attention_dropout_rate: 0.1

model_conf:
ctc_weight: 0.3
lsm_weight: 0.1
interctc_weight: 0.5
length_normalized_loss: false
extract_feats_in_collect_stats: false

optim: adam
optim_conf:
lr: 0.002
scheduler: warmuplr
scheduler_conf:
warmup_steps: 25000

specaug: specaug
specaug_conf:
apply_time_warp: true
time_warp_window: 5
time_warp_mode: bicubic
apply_freq_mask: true
freq_mask_width_range:
- 0
- 30
num_freq_mask: 2
apply_time_mask: true
time_mask_width_range:
- 0
- 40
num_time_mask: 2

freeze_param: [
"frontend.upstream"
]

frontend: s3prl
frontend_conf:
frontend_conf:
upstream: wav2vec2_url # Note: If the upstream is changed, please change the input_size in the preencoder.
path_or_url: https://huggingface.co/s3prl/converted_ckpts/resolve/main/xlsr2_300m.pt
download_dir: ./hub
multilayer_feature: True

preencoder: linear
preencoder_conf:
input_size: 1024 # Note: If the upstream is changed, please change this value accordingly.
output_size: 80