# Audio classification examples The following examples showcase how to fine-tune `HuBERT` for audio classification using PyTorch. Speech recognition models that have been pretrained in unsupervised fashion on audio data alone, *e.g.* [Wav2Vec2](https://huggingface.co/transformers/master/model_doc/wav2vec2.html), [HuBERT](https://huggingface.co/transformers/master/model_doc/hubert.html), [XLSR-Wav2Vec2](https://huggingface.co/transformers/master/model_doc/xlsr_wav2vec2.html), have shown to require only very little annotated data to yield good performance on speech classification datasets. ## SUPERB Dataset The following command shows how to fine-tune [hubert-base-ls960](https://huggingface.co/facebook/hubert-base-ls960) on the 🗣️ [Keyword Spotting subset](https://huggingface.co/datasets/superb#ks) of the SUPERB dataset. ```bash python examples/audio-classification/run_audio_classification.py \ --model_name_or_path facebook/hubert-base-ls960 \ --ipu_config_name Graphcore/hubert-base-ipu \ --dataset_name superb \ --dataset_config_name ks \ --output_dir hubert-base-superb-ks \ --overwrite_output_dir \ --remove_unused_columns False \ --do_train \ --do_eval \ --learning_rate 3e-5 \ --max_length_seconds 1 \ --attention_mask False \ --warmup_ratio 0.1 \ --lr_schedule linear \ --weight_decay 0.01 \ --num_train_epochs 5 \ --per_device_train_batch_size 2 \ --gradient_accumulation_steps 16 \ --per_device_eval_batch_size 8 \ --n_ipu 16 \ --dataloader_num_workers 64 \ --dataloader_drop_last True \ --logging_strategy steps \ --logging_steps 10 \ --seed 0 ``` ## Common Language The following command shows how to fine-tune [hubert-base-ls960](https://huggingface.co/facebook/hubert-base-ls960) for 🌎 **Language Identification** on the [CommonLanguage dataset](https://huggingface.co/datasets/anton-l/common_language). ```bash python examples/audio-classification/run_audio_classification.py \ --model_name_or_path facebook/hubert-base-ls960 \ --ipu_config_name Graphcore/hubert-base-ipu \ --dataset_name common_language \ --audio_column_name audio \ --label_column_name language \ --output_dir /tmp/hubert-base-common-language \ --overwrite_output_dir \ --remove_unused_columns False \ --do_train \ --do_eval \ --max_length_seconds 13 \ --attention_mask False \ --learning_rate 1e-4 \ --warmup_ratio 0.25 \ --lr_schedule linear \ --num_train_epochs 10 \ --per_device_train_batch_size 1 \ --gradient_accumulation_steps 32 \ --per_device_eval_batch_size 4 \ --dataloader_num_workers 64 \ --dataloader_drop_last True \ --n_ipu 16 \ --logging_strategy steps \ --logging_steps 10 \ --seed 0 ``` ## Sharing your model on 🤗 Hub 0. If you haven't already, [sign up](https://huggingface.co/join) for a 🤗 account 1. Make sure you have `git-lfs` installed and git set up. ```bash $ apt install git-lfs ``` 2. Log in with your HuggingFace account credentials using `huggingface-cli` ```bash $ huggingface-cli login # ...follow the prompts ``` 3. When running the script, pass the following arguments: ```bash python run_audio_classification.py \ --push_to_hub \ --hub_model_id \ ... ```