ASR_verbatim_v1 / run-subs.sh
Jakob Poncelet
First model version
e2baad4
raw
history blame
3.97 kB
#!/usr/bin/env bash
export PYTHONPATH=""
source /esat/spchtemp/scratch/jponcele/anaconda3/bin/activate espnet2
python --version
# Set bash to 'debug' mode, it will exit on :
# -e 'error', -u 'undefined variable', -o ... 'error in pipeline', -x 'print commands',
set -e
set -u
set -o pipefail
####################################################
stage=11
stop_stage=11
####################################################
# notes: geen speed perturbation, geen LM, geen word LM, geen NGRAM LM
# EXP
outdir=/esat/spchtemp/scratch/jponcele/espnet2
expdir=${outdir}/exp/exp-subs-chained
st_tag=train_subtitling_chained_PL_C8_new
st_stats_dir=/esat/spchtemp/scratch/jponcele/espnet2/exp/exp-st/st_stats_fbank_pitch_vl_joint_bpe5000
nj=4
ngpu=1
# DATA
st_train_set=st_train
st_valid_set=st_valid
st_test_set="subs_annot"
asr_train_set=train_s
asr_valid_set=valid_s
asr_test_set=dev_s
subs_train_set=subs_train
subs_valid_set=subs_valid
subs_test_set=subs_test
traincomps="a;b;c;d;f;g;h;i;j;k;l;m;n;o"
decodecomps="b;f;g;h;i;j;k;l;m;n;o"
local_data_opts="--repstr false --lowercase true --outdir data --traincomps ${traincomps} --decodecomps ${decodecomps}"
subs_dir=/users/spraak/jponcele/vrt-scraper/vrtnew_subtitles_4feb
local_subs_opts="--outdir data --subsdir ${subs_dir}"
feats_type=fbank_pitch
# LM
use_word_lm=false # not yet supported!
use_lm=false
lm_config=conf/train_lm_transformer.yaml
use_ngram=false
# ST
feats_normalize=utterance_mvn # recommended for pretrained models instead of globalmvn
st_config=conf/tuning/train_subtitling_chained_C8_new.yaml
inference_config=conf/st_decode_chained.yaml
inference_nj=64
inference_st_model=averaged_model_81epochs.pth #valid.acc_asr.ave.pth
st_args="--batch_type custom_folded --valid_batch_type custom_folded" # "--input_size 0" # to use raw audio for w2v2 encoder
./subs.sh \
--stage ${stage} \
--stop_stage ${stop_stage} \
--ngpu ${ngpu} \
--nj ${nj} \
--gpu_inference false \
--dumpdir ${outdir}/dump \
--expdir ${expdir} \
--feats_type ${feats_type} \
--audio_format wav \
--min_wav_duration 0.1 \
--max_wav_duration 30 \
--token_joint true \
--src_token_type bpe \
--src_nbpe 5000 \
--src_bpemode unigram \
--src_case lc \
--tgt_token_type bpe \
--tgt_nbpe 5000 \
--tgt_bpemode unigram \
--tgt_case lc \
--oov "<unk>" \
--lang "vl" \
--src_lang "verbatim" \
--tgt_lang "subtitle" \
--local_subs_opts "${local_subs_opts}" \
--local_data_opts "${local_data_opts}" \
--use_lm ${use_lm} \
--use_word_lm ${use_word_lm} \
--lm_config ${lm_config} \
--use_ngram ${use_ngram} \
--st_config ${st_config} \
--st_args "${st_args}" \
--st_tag ${st_tag} \
--inference_config ${inference_config} \
--inference_nj ${inference_nj} \
--feats_normalize ${feats_normalize} \
--st_train_set "${st_train_set}" \
--st_valid_set "${st_valid_set}" \
--st_test_set "${st_test_set}" \
--asr_train_set ${asr_train_set} \
--asr_valid_set ${asr_valid_set} \
--asr_test_set ${asr_test_set} \
--subs_train_set ${subs_train_set} \
--subs_valid_set ${subs_valid_set} \
--subs_test_set ${subs_test_set} \
--st_stats_dir ${st_stats_dir} \
--inference_st_model ${inference_st_model} \
# --pretrained_asr ${pretrained_asr} \
#train_set=train_si284
#valid_set=test_dev93
#test_sets="test_dev93 test_eval92"
#
#./asr.sh \
# --lang "en" \
# --use_lm true \
# --token_type char \
# --nbpe 80 \
# --nlsyms_txt data/nlsyms.txt \
# --lm_config conf/train_lm_transformer.yaml \
# --asr_config conf/train_asr_transformer.yaml \
# --inference_config conf/decode.yaml \
# --train_set "${train_set}" \
# --valid_set "${valid_set}" \
# --test_sets "${test_sets}" \
# --bpe_train_text "data/train_si284/text" \
# --lm_train_text "data/train_si284/text data/local/other_text/text" "$@"