Closed csukuangfj closed 1 month ago
The corpus can be downloaded from https://huggingface.co/datasets/amphion/Emilia-Dataset
(py38) kuangfangjun:recipes$ mkdir ~/t5 (py38) kuangfangjun:recipes$ time lhotse prepare emilia --num-jobs 30 --lang de /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia /star-fj/fa ngjun/t5/ Processing /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia/raw/DE/DE_B00000.jsonl with 30 jobs: 166901it [00:41, 4024.66it/s] Processing /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia/raw/DE/DE_B00001.jsonl with 30 jobs: 152274it [00:33, 4592.59it/s] Processing /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia/raw/DE/DE_B00002.jsonl with 30 jobs: 167050it [00:49, 3391.75it/s] Processing /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia/raw/DE/DE_B00003.jsonl with 30 jobs: 45364it [00:09, 4747.71it/s] Processing /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia/raw/DE/DE_B00004.jsonl with 30 jobs: 48257it [00:35, 1378.46it/s] Processing /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia/raw/DE/DE_B00005.jsonl with 30 jobs: 32706it [00:02, 13124.74it/s] Processing /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia/raw/DE/DE_B00006.jsonl with 30 jobs: 22127it [00:05, 4183.53it/s] Processing /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia/raw/DE/DE_B00007.jsonl with 30 jobs: 14712it [00:01, 13699.85it/s] Processing /star-fj/fangjun/data/tts/emilia/download/Amphion___Emilia/raw/DE/DE_B00008.jsonl with 30 jobs: 3718it [00:00, 13919.69it/s] Collecting futures: 100%|_________________________________________________________________________________| 653109/653109 [00:04<00:00, 144918.00it/s] real 4m15.395s user 4m30.221s sys 0m57.310s
(Note: Only audio files of DE_B00000.jsonl and DE_B00001.jsonl in the above example are extracted to the corpus directory).
DE_B00000.jsonl
DE_B00001.jsonl
Generated files are
(py38) kuangfangjun:recipes$ cd ~/t5 (py38) kuangfangjun:t5$ ls -lh total 30M -rw-r--r-- 1 kuangfangjun root 30M Oct 21 15:57 emilia_cuts_DE.jsonl.gz
(py38) kuangfangjun:t5$ lhotse cut describe ./emilia_cuts_DE.jsonl.gz
prints
Cut statistics: _________________________________________ _ Cuts count: _ 319175 _ _________________________________________ _ Total duration (hh:mm:ss) _ 766:12:03 _ _________________________________________ _ mean _ 8.6 _ _________________________________________ _ std _ 5.0 _ _________________________________________ _ min _ 3.0 _ _________________________________________ _ 25% _ 4.7 _ _________________________________________ _ 50% _ 7.2 _ _________________________________________ _ 75% _ 11.3 _ _________________________________________ _ 99% _ 24.2 _ _________________________________________ _ 99.5% _ 28.0 _ _________________________________________ _ 99.9% _ 29.6 _ _________________________________________ _ max _ 30.0 _ _________________________________________ _ Recordings available: _ 319175 _ _________________________________________ _ Features available: _ 0 _ _________________________________________ _ Supervisions available: _ 319175 _ _________________________________________ SUPERVISION custom fields: - dnsmos (in 319175 cuts) Speech duration statistics: ___________________________________________________________________ _ Total speech duration _ 766:12:03 _ 100.00% of recording _ ___________________________________________________________________ _ Total speaking time duration _ 766:12:04 _ 100.00% of recording _ ___________________________________________________________________ _ Total silence duration _ 00:00:01 _ 0.00% of recording _ ___________________________________________________________________
The corpus can be downloaded from https://huggingface.co/datasets/amphion/Emilia-Dataset
Usage example
(Note: Only audio files of
DE_B00000.jsonlandDE_B00001.jsonlin the above example are extracted to the corpus directory).Generated files are
prints