forked from espnet/espnet
-
Notifications
You must be signed in to change notification settings - Fork 1
/
test_integration_espnet2.sh
executable file
·225 lines (200 loc) · 12.1 KB
/
test_integration_espnet2.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
#!/usr/bin/env bash
set -euo pipefail
source tools/activate_python.sh
PYTHONPATH="${PYTHONPATH:-}:$(pwd)/tools/s3prl"
export PYTHONPATH
python="coverage run --append"
cwd=$(pwd)
#### Make sure chainer-independent ####
python3 -m pip uninstall -y chainer
# [ESPnet2] test asr recipe
cd ./egs2/mini_an4/asr1
echo "==== [ESPnet2] ASR ==="
./run.sh --stage 1 --stop-stage 1
feats_types="raw fbank_pitch"
token_types="bpe char"
for t in ${feats_types}; do
./run.sh --stage 2 --stop-stage 4 --feats-type "${t}" --python "${python}"
done
for t in ${token_types}; do
./run.sh --stage 5 --stop-stage 5 --token-type "${t}" --python "${python}"
done
for t in ${feats_types}; do
for t2 in ${token_types}; do
echo "==== feats_type=${t}, token_types=${t2} ==="
./run.sh --ngpu 0 --stage 6 --stop-stage 13 --skip-upload false --feats-type "${t}" --token-type "${t2}" \
--asr-args "--max_epoch=1" --lm-args "--max_epoch=1" --python "${python}"
done
done
echo "==== feats_type=raw, token_types=bpe, model_conf.extract_feats_in_collect_stats=False, normalize=utt_mvn ==="
./run.sh --ngpu 0 --stage 10 --stop-stage 13 --skip-upload false --feats-type "raw" --token-type "bpe" \
--feats_normalize "utterance_mvn" --lm-args "--max_epoch=1" --python "${python}" \
--asr-args "--model_conf extract_feats_in_collect_stats=false --max_epoch=1"
echo "==== use_streaming, feats_type=raw, token_types=bpe, model_conf.extract_feats_in_collect_stats=False, normalize=utt_mvn ==="
./run.sh --use_streaming true --ngpu 0 --stage 6 --stop-stage 13 --skip-upload false --feats-type "raw" --token-type "bpe" \
--feats_normalize "utterance_mvn" --lm-args "--max_epoch=1" --python "${python}" \
--asr-args "--model_conf extract_feats_in_collect_stats=false --max_epoch=1 --encoder=contextual_block_transformer --decoder=transformer
--encoder_conf block_size=40 --encoder_conf hop_size=16 --encoder_conf look_ahead=16"
if python3 -c "import k2" &> /dev/null; then
echo "==== use_k2, num_paths > nll_batch_size, feats_type=raw, token_types=bpe, model_conf.extract_feats_in_collect_stats=False, normalize=utt_mvn ==="
./run.sh --num_paths 500 --nll_batch_size 20 --use_k2 true --ngpu 0 --stage 12 --stop-stage 13 --skip-upload false --feats-type "raw" --token-type "bpe" \
--feats_normalize "utterance_mvn" --lm-args "--max_epoch=1" --python "${python}" \
--asr-args "--model_conf extract_feats_in_collect_stats=false --max_epoch=1"
echo "==== use_k2, num_paths == nll_batch_size, feats_type=raw, token_types=bpe, model_conf.extract_feats_in_collect_stats=False, normalize=utt_mvn ==="
./run.sh --num_paths 20 --nll_batch_size 20 --use_k2 true --ngpu 0 --stage 12 --stop-stage 13 --skip-upload false --feats-type "raw" --token-type "bpe" \
--feats_normalize "utterance_mvn" --lm-args "--max_epoch=1" --python "${python}" \
--asr-args "--model_conf extract_feats_in_collect_stats=false --max_epoch=1"
fi
if python3 -c "from warprnnt_pytorch import RNNTLoss" &> /dev/null; then
echo "==== [ESPnet2] ASR Transducer (standalone) ==="
for t in ${token_types}; do
asr_tag="transducer_${t}"
echo "==== [Conformer-RNN-T] feats_type=raw, token_types=${t}, model_conf.extract_feats_in_collect_stats=False, normalize=utt_mvn ==="
./run.sh --asr_task "asr_transducer" --ngpu 0 --stage 10 --stop-stage 13 --skip-upload false --feats-type "raw" --token-type ${t} \
--feats_normalize "utterance_mvn" --lm-args "--max_epoch=1" --python "${python}" --inference_asr_model "valid.loss.best.pth" \
--asr-tag "${asr_tag}_conformer" --asr-args "--model_conf extract_feats_in_collect_stats=false --max_epoch=1 \
--encoder_conf body_conf='[{'block_type': 'conformer', 'hidden_size': 30, 'linear_size': 30, 'heads': 2, 'conv_mod_kernel_size': 3}]' \
--decoder_conf='{'embed_size': 30, 'hidden_size': 30}' --joint_network_conf joint_space_size=30"
echo "==== [Streaming Conformer-RNN-T] feats_type=raw, token_types=${t}, model_conf.extract_feats_in_collect_stats=False, normalize=utt_mvn ==="
./run.sh --asr_task "asr_transducer" --ngpu 0 --stage 10 --stop-stage 13 --skip-upload false --feats-type "raw" --token-type ${t} \
--feats_normalize "utterance_mvn" --lm-args "--max_epoch=1" --python "${python}" --inference_asr_model "valid.loss.best.pth" \
--asr-tag "${asr_tag}_conformer_streaming" --asr-args "--model_conf extract_feats_in_collect_stats=false --max_epoch=1 \
--encoder_conf main_conf='{'dynamic_chunk_training': True}' \
--encoder_conf body_conf='[{'block_type': 'conformer', 'hidden_size': 30, 'linear_size': 30, 'heads': 2, 'conv_mod_kernel_size': 3}]' \
--decoder_conf='{'embed_size': 30, 'hidden_size': 30}' --joint_network_conf joint_space_size=30 " \
--inference-args "--streaming true --chunk_size 2 --left_context 2 --right_context 0"
done
fi
# Remove generated files in order to reduce the disk usage
rm -rf exp dump data
cd "${cwd}"
# [ESPnet2] test tts recipe
cd ./egs2/mini_an4/tts1
echo "==== [ESPnet2] TTS ==="
./run.sh --ngpu 0 --stage 1 --stop-stage 8 --skip-upload false --train-args "--max_epoch 1" --python "${python}"
# Remove generated files in order to reduce the disk usage
rm -rf exp dump data
# [ESPnet2] test gan-tts recipe
# NOTE(kan-bayashi): pytorch 1.4 - 1.6 works but 1.6 has a problem with CPU,
# so we test this recipe using only pytorch > 1.6 here.
# See also: https://github.com/pytorch/pytorch/issues/42446
if python3 -c 'import torch as t; from packaging.version import parse as L; assert L(t.__version__) > L("1.6")' &> /dev/null; then
./run.sh --fs 22050 --tts_task gan_tts --feats_extract linear_spectrogram --feats_normalize none --inference_model latest.pth \
--ngpu 0 --stop-stage 8 --skip-upload false --train-args "--num_iters_per_epoch 1 --max_epoch 1" --python "${python}"
rm -rf exp dump data
fi
cd "${cwd}"
# [ESPnet2] test enh recipe
if python -c 'import torch as t; from packaging.version import parse as L; assert L(t.__version__) >= L("1.2.0")' &> /dev/null; then
cd ./egs2/mini_an4/enh1
echo "==== [ESPnet2] ENH ==="
./run.sh --stage 1 --stop-stage 1 --python "${python}"
feats_types="raw"
for t in ${feats_types}; do
echo "==== feats_type=${t} ==="
./run.sh --ngpu 0 --stage 2 --stop-stage 10 --skip-upload false --feats-type "${t}" --spk-num 1 --enh-args "--max_epoch=1" --python "${python}"
./run.sh --ngpu 0 --stage 2 --stop-stage 10 --skip-upload false --feats-type "${t}" --spk-num 1 --enh-args "--max_epoch=1" --python "${python}" --use_preprocessor true --extra_wav_list "rirs.scp noises.scp" --enh_config ./conf/train_with_preprocessor.yaml
./run.sh --ngpu 0 --stage 2 --stop-stage 10 --skip-upload false --feats-type "${t}" --spk-num 1 --enh-args "--max_epoch=1" --python "${python}" --enh_config conf/train_with_dynamic_mixing.yaml --dynamic_mixing true --spk-num 2
done
# Remove generated files in order to reduce the disk usage
rm -rf exp dump data
cd "${cwd}"
fi
# [ESPnet2] test ssl1 recipe
if python3 -c "import fairseq" &> /dev/null; then
cd ./egs2/mini_an4/ssl1
echo "==== [ESPnet2] SSL1/HUBERT ==="
./run.sh --ngpu 0 --stage 1 --stop-stage 7 --feats-type "raw" --token_type "word" --skip-upload false --pt-args "--max_epoch=1" --pretrain_start_iter 0 --pretrain_stop_iter 1 --python "${python}"
# Remove generated files in order to reduce the disk usage
rm -rf exp dump data
cd "${cwd}"
fi
# [ESPnet2] test enh_asr1 recipe
if python -c 'import torch as t; from packaging.version import parse as L; assert L(t.__version__) >= L("1.2.0")' &> /dev/null; then
cd ./egs2/mini_an4/enh_asr1
echo "==== [ESPnet2] ENH_ASR ==="
./run.sh --ngpu 0 --stage 0 --stop-stage 15 --skip-upload_hf false --feats-type "raw" --spk-num 1 --enh_asr_args "--max_epoch=1 --enh_separator_conf num_spk=1" --python "${python}"
# Remove generated files in order to reduce the disk usage
rm -rf exp dump data
cd "${cwd}"
fi
# [ESPnet2] test st recipe
cd ./egs2/mini_an4/st1
echo "==== [ESPnet2] ST ==="
./run.sh --stage 1 --stop-stage 1
feats_types="raw fbank_pitch"
token_types="bpe char"
for t in ${feats_types}; do
./run.sh --stage 2 --stop-stage 4 --feats-type "${t}" --python "${python}"
done
for t in ${token_types}; do
./run.sh --stage 5 --stop-stage 5 --tgt_token_type "${t}" --src_token_type "${t}" --python "${python}"
done
for t in ${feats_types}; do
for t2 in ${token_types}; do
echo "==== feats_type=${t}, token_types=${t2} ==="
./run.sh --ngpu 0 --stage 6 --stop-stage 13 --skip-upload false --feats-type "${t}" --tgt_token_type "${t2}" --src_token_type "${t2}" \
--st-args "--max_epoch=1" --lm-args "--max_epoch=1" --inference_args "--beam_size 5" --python "${python}"
done
done
echo "==== feats_type=raw, token_types=bpe, model_conf.extract_feats_in_collect_stats=False, normalize=utt_mvn ==="
./run.sh --ngpu 0 --stage 10 --stop-stage 13 --skip-upload false --feats-type "raw" --tgt_token_type "bpe" --src_token_type "bpe" \
--feats_normalize "utterance_mvn" --lm-args "--max_epoch=1" --inference_args "--beam_size 5" --python "${python}" \
--st-args "--model_conf extract_feats_in_collect_stats=false --max_epoch=1"
echo "==== use_streaming, feats_type=raw, token_types=bpe, model_conf.extract_feats_in_collect_stats=False, normalize=utt_mvn ==="
./run.sh --use_streaming true --ngpu 0 --stage 6 --stop-stage 13 --skip-upload false --feats-type "raw" --tgt_token_type "bpe" --src_token_type "bpe" \
--feats_normalize "utterance_mvn" --lm-args "--max_epoch=1" --inference_args "--beam_size 5" --python "${python}" \
--st-args "--model_conf extract_feats_in_collect_stats=false --max_epoch=1 --encoder=contextual_block_transformer --decoder=transformer
--encoder_conf block_size=40 --encoder_conf hop_size=16 --encoder_conf look_ahead=16"
# Remove generated files in order to reduce the disk usage
rm -rf exp dump data
cd "${cwd}"
# [ESPnet2] Validate configuration files
echo "<blank>" > dummy_token_list
echo "==== [ESPnet2] Validation configuration files ==="
if python3 -c 'import torch as t; from packaging.version import parse as L; assert L(t.__version__) >= L("1.8.0")' &> /dev/null; then
for f in egs2/*/asr1/conf/train_asr*.yaml; do
if [ "$f" == "egs2/fsc/asr1/conf/train_asr.yaml" ]; then
if ! python3 -c "import s3prl" > /dev/null; then
continue
fi
fi
${python} -m espnet2.bin.asr_train --config "${f}" --iterator_type none --dry_run true --output_dir out --token_list dummy_token_list
done
for f in egs2/*/asr1/conf/train_lm*.yaml; do
${python} -m espnet2.bin.lm_train --config "${f}" --iterator_type none --dry_run true --output_dir out --token_list dummy_token_list
done
for f in egs2/*/tts1/conf/train*.yaml; do
${python} -m espnet2.bin.tts_train --config "${f}" --iterator_type none --normalize none --dry_run true --output_dir out --token_list dummy_token_list
done
for f in egs2/*/enh1/conf/train*.yaml; do
${python} -m espnet2.bin.enh_train --config "${f}" --iterator_type none --dry_run true --output_dir out
done
for f in egs2/*/ssl1/conf/train*.yaml; do
${python} -m espnet2.bin.hubert_train --config "${f}" --iterator_type none --normalize none --dry_run true --output_dir out --token_list dummy_token_list
done
for f in egs2/*/enh_asr1/conf/train_enh_asr*.yaml; do
${python} -m espnet2.bin.enh_s2t_train --config "${f}" --iterator_type none --dry_run true --output_dir out --token_list dummy_token_list
done
fi
# These files must be same each other.
for base in cmd.sh conf/slurm.conf conf/queue.conf conf/pbs.conf; do
file1=
for f in egs2/*/*/"${base}"; do
if [ -z "${file1}" ]; then
file1="${f}"
fi
diff "${file1}" "${f}" || { echo "Error: ${file1} and ${f} differ: To solve: for f in egs2/*/*/${base}; do cp egs2/TEMPLATE/asr1/${base} \${f}; done" ; exit 1; }
done
done
echo "==== [ESPnet2] test setup.sh ==="
for d in egs2/TEMPLATE/*; do
if [ -d "${d}" ]; then
d="${d##*/}"
egs2/TEMPLATE/"$d"/setup.sh egs2/test/"${d}"
fi
done
echo "=== report ==="
coverage combine egs2/*/*/.coverage
coverage report
coverage xml