speech-to-text-semisupervised ============================= Pseudolabel Nusantara Audiobook using Whisper Large V3 ------------------------------------------------------ download ~~~~~~~~ All data uploaded at https://huggingface.co/datasets/mesolitica/nusantara-audiobook Nusantara Audiobook ------------------- This directory to gather semisupervised transcribed on Malay audiobook. :: All the videos, songs, images, and graphics used in the video belong to their respective owners and I does not claim any right over them. Copyright Disclaimer under section 107 of the Copyright Act of 1976, allowance is made for "fair use" for purposes such as criticism, comment, news reporting, teaching, scholarship, education and research. Fair use is a use permitted by copyright statute that might otherwise be infringing. Download ~~~~~~~~ 1. https://f000.backblazeb2.com/file/malaya-speech-model/data/dari-pasentran-ke-istana.gz - Originally from https://nusantaraudiobooks.com/books/dari-pesantren-ke-istana-biografi-presiden-ke-4-indonesia-kiai-haji-abdurrahman-wahid - 44100 sample rate, super clean. - narrator Danny Abdullah. - approximate 19.63 hours. - VAD = 2, https://f000.backblazeb2.com/file/malaya-speech-model/data/dari-pasentran-ke-istana-short.gz - Semisupervised using PyTorch Conformer Medium, `semisupervised-pasentran-turki.json `__, notebook `semisupervised-pasentran-turki.ipynb `__. - Put commas and apply true case, `true-case-pasentran-turki.json `__ notebook `put-comma-true-case-pasentran-turki.ipynb `__ 2. https://f000.backblazeb2.com/file/malaya-speech-model/data/turki.gz - Originally from https://nusantaraudiobooks.com/books/dari-sultan-hingga-ataturk-turki - 44100 sample rate, super clean. - narrator Danny Abdullah. - approximate 7.73 hours. - VAD = 2, https://f000.backblazeb2.com/file/malaya-speech-model/data/turki-short.gz - Semisupervised using PyTorch Conformer Medium, `semisupervised-pasentran-turki.json `__, notebook `semisupervised-pasentran-turki.ipynb `__. - Put commas and apply true case, `true-case-pasentran-turki.json `__ notebook `put-comma-true-case-pasentran-turki.ipynb `__ 3. https://f000.backblazeb2.com/file/malaya-speech-model/data/salina.gz - Originally from https://nusantaraudiobooks.com/books/salina - 44100 sample rate, super clean. - narrator T Elida Bustaman. - approximate 24.66 hours. - VAD = 2, https://f000.backblazeb2.com/file/malaya-speech-model/data/salina-short.gz - Semisupervised using PyTorch Conformer Medium, `semisupervised-salina.json `__, notebook `semisupervised-salina.ipynb `__. - Put commas and apply true case, `true-case-salina.json `__ notebook `put-comma-true-case-salina.ipynb `__ 4. Text only, https://f000.backblazeb2.com/file/malaya-speech-model/data/text-audiobook.tar.gz 5. Test set, https://f000.backblazeb2.com/file/malaya-speech-model/data/testset-audiobook.tar.gz 6. Train set with augmentation, https://f000.backblazeb2.com/file/malaya-speech-model/data/trainset-audiobook.tar.gz 7. https://f000.backblazeb2.com/file/malaya-speech-model/data/salina-supervised-sani.tar.gz - Originally from https://nusantaraudiobooks.com/books/salina - Supervised by https://github.com/khursani8 - approximate 19.32 hours. - Put commas and apply true case, `comma-salina-sani.json `__, notebook `alignment-salina-sani.ipynb `__. 8. https://f000.backblazeb2.com/file/malaya-speech-model/data/dari-pasentran-ke-istana-supervised-sani.tar.gz - Originally from https://nusantaraudiobooks.com/books/dari-pesantren-ke-istana-biografi-presiden-ke-4-indonesia-kiai-haji-abdurrahman-wahid - Supervised by https://github.com/khursani8 - approximate 13.84 hours. - Put commas and apply true case, `comma-dari-pasentran-ke-istana-sani.json `__, notebook `alignment-dari-pasentran-ke-istana-sani.ipynb `__. Citation ~~~~~~~~ .. code:: bibtex @misc{Malay-Dataset, We gather Bahasa Malaysia corpus!, Semisupervised Speech Recognition from Audiobook, author = {Husein, Zolkepli}, title = {Malay-Dataset}, year = {2018}, publisher = {GitHub}, journal = {GitHub repository}, howpublished = {\url{https://github.com/huseinzol05/malaya-speech/tree/master/data/semisupervised-audiobook}} } distributed multi-GPUs pseudolabel using Whisper on Malaya-Speech STT --------------------------------------------------------------------- This pseudolabel included fast hashing load audio files and continue last step decoded. download ~~~~~~~~ All data uploaded at https://huggingface.co/datasets/mesolitica/pseudolabel-malaysian-youtube-whisper-large-v3 how-to ~~~~~~ 1. Generate chunks hash map, `generate-global-indices.ipynb `__. Use torchrun ^^^^^^^^^^^^ .. code:: bash NCCL_P2P_DISABLE=1 NCCL_IB_DISABLE=1 ~/.local/bin/torchrun --nproc_per_node 2 \ -m run \ --indices_filename=indices-crawl-malaya-speech.json --batch_size=16 NCCL is not required. distributed multi-GPUs pseudolabel using Whisper on Malaysian Youtube videos ---------------------------------------------------------------------------- This pseudolabel included fast hashing load audio files and continue last step decoded. download ~~~~~~~~ All data uploaded at https://huggingface.co/datasets/mesolitica/pseudolabel-malaysian-youtube-whisper-large-v3 how-to ~~~~~~ 1. Prepare chunks hash map, `prepare-indices-chunks.ipynb `__. 2. Generate chunks hash map, `generate-global-indices.ipynb `__. Use accelerate ^^^^^^^^^^^^^^ 1. Configure accelerate, .. code:: bash accelerate config 2. Run accelerate, .. code:: bash ~/my-env/bin/accelerate launch run.py --indices_filename=global-indices.json --batch_size=4 Use torchrun ^^^^^^^^^^^^ .. code:: bash NCCL_P2P_DISABLE=1 NCCL_IB_DISABLE=1 ~/my-env/bin/torchrun --nproc_per_node 2 \ -m run \ --indices_filename=global-indices.json --batch_size=4 NCCL is not required. Run in 4x A100 """""""""""""" We use batch size of 52, .. code:: bash NCCL_P2P_DISABLE=1 NCCL_IB_DISABLE=1 torchrun --nproc_per_node 4 \ -m run \ --indices_filename=crawl-youtube-global-indices.json --batch_size=52 Predict language using Speechbrain ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ .. code:: bash NCCL_P2P_DISABLE=1 NCCL_IB_DISABLE=1 torchrun --nproc_per_node 4 \ -m run-predict-lang \ --batch_size=32 Noisy Audiobook --------------- :: All the videos, songs, images, and graphics used in the video belong to their respective owners and I does not claim any right over them. Copyright Disclaimer under section 107 of the Copyright Act of 1976, allowance is made for "fair use" for purposes such as criticism, comment, news reporting, teaching, scholarship, education and research. Fair use is a use permitted by copyright statute that might otherwise be infringing. download ~~~~~~~~ 1. Harry Potter, - audio, https://huggingface.co/datasets/mesolitica/semisupervised-audiobook/resolve/main/harry-potter-noisy.tar.gz - transcription, https://huggingface.co/datasets/mesolitica/semisupervised-audiobook/resolve/main/harry-potter-processed.json 2. Teme, - audio, https://huggingface.co/datasets/mesolitica/semisupervised-audiobook/resolve/main/teme-noisy.tar.gz - transcription, https://huggingface.co/datasets/mesolitica/semisupervised-audiobook/resolve/main/teme-processed.json 3. Bukan Kerana Aku, - audio, https://huggingface.co/datasets/mesolitica/semisupervised-audiobook/resolve/main/bukan-kerana-aku-noisy.tar.gz - transcription, https://huggingface.co/datasets/mesolitica/semisupervised-audiobook/resolve/main/bukan-kerana-aku-processed.json