GitHub - Chung-I/s5-taibun-aug

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
conf		conf
local		local
steps		steps
tsm		tsm
utils		utils
.gitignore		.gitignore
cmd.sh		cmd.sh
decode_all.sh		decode_all.sh
decode_mandarin.sh		decode_mandarin.sh
decode_taibun.sh		decode_taibun.sh
decode_tailo_ftv.sh		decode_tailo_ftv.sh
dict_seg.py		dict_seg.py
draw_tree.sh		draw_tree.sh
format_accuracy_for_plot.sh		format_accuracy_for_plot.sh
get_phns.sh		get_phns.sh
make_tailo_dict.py		make_tailo_dict.py
pack_model.sh		pack_model.sh
pack_server.sh		pack_server.sh
path.sh		path.sh
plot_accuracy.py		plot_accuracy.py
prons_to_lexicon.sh		prons_to_lexicon.sh
readme.txt		readme.txt
run.sh		run.sh
test_bat.sh		test_bat.sh
text_prepare_tailo.py		text_prepare_tailo.py

Repository files navigation

data augmentation needs musan (http://www.openslr.org/17/).
set $CORPUS_DIR in path.sh and put musan under $CORPUS_DIR. (see local/nnet3/run_aug.sh).
reverberation needs RIRS_NOISES (https://www.openslr.org/28/).
put RIRS_NOISES here. (see local/nnet3/run_aug.sh)

python package requirements:
ckiptagger
regex
zhon
cn2an
opencc

to prepare PTS_TW-extra data, run:
mandarin:
local/prepare_pts_data.sh --train-dir PTS_TW-extra --data-dir data/<dataset_name> --add-parent-prefix true \
  --lexicon-path language/mandarin_phn_lexiconp.txt --txtdir mandarin_text

taibun:
local/prepare_pts_data.sh --train-dir PTS_TW-extra --data-dir data/<dataset_name> --add-parent-prefix true \
  --lexicon-path language/hanlo_tailo_phn_lexiconp.txt --txtdir taibun_text

上次的PTS_TW-extra的text是mandarin跟taibun在同一個檔案裡，
這次給的PTS_TW-extra-textonly只是把它分開來成taibun_text跟mandarin_text，作為--txtdir的參數