Transferability of Speech Separation with Meta-learning.
Scripts to create the testing set speech mixtures.
step 1: Download languages zh-CN, zh-TW, zh-HK from https://commonvoice.mozilla.org/zh-TW/datasets.
step 2: Convert mp3 to wav
python3 mp3_to_wav.py --data_dir /path/to/cv-corpus-6.1-2020-12-11/ \
--output_dir /path/to/common-voice-zh-split-4s \
--lang [lang] \
--split test
step 3: Generate mix list
2-speaker
python3 mix_list_gen_cv_zh.py --data_dir /path/to/common-voice-zh-split-4s \
--lang [lang] \
--do_tt
3-speaker
python3 mix_list_gen_cv_zh_3spk.py --data_dir /path/to/common-voice-zh-split-4s \
--lang [lang] \
--do_tt
step 4: Generate mixtures according to mix list
cd create-speaker-mixtures_cv_zh
octave
pkg load signal
create_wav_2speakers
create_wav_3speakers
repeat step 3 and 4 for each language
Download Speech Accent Archive mp3 files from https://www.kaggle.com/rtatman/speech-accent-archive.
split audio and convert mp3 to wave
python3 split_SAA_audio.py --data_dir /path/to/recordings --output_dir /path/to/multi_accent_split_4s/
Generate mix list
2-speaker
python3 mix_list_gen_mac_2spk_task_English.py --data_dir /path/to/multi_accent_split_4s/ --do_tr --do_cv --do_tt
3-speaker
python3 mix_list_gen_mac_3spk_task_English.py --data_dir /path/to/multi_accent_split_4s/ --do_tr --do_cv --do_tt
Generate mixtures according to mix list
cd create-speaker-mixtures_mac_English
octave
pkg load signal
create_wav_2speakers
create_wav_3speakers