wav2vec2-base_down_on

This model is a fine-tuned version of facebook/wav2vec2-base on the MatsRooth/down_on dataset. It achieves the following results on the evaluation set:

• Loss: 0.1385
• Accuracy: 0.9962

Model description

Binary classifier using facebook/wav2vec2/base for the words "down" and "on".

Intended uses & limitations

This is a demo of binary audio classification that illustrates data layout, training and evaluation using python and slurm.

Training and evaluation data

The data are utterances of "down" and "on" in superb ks. See down_on_copy.py for the subsetting. This puts wav files in locations like down_on/data/train/on and down_on/data/train/down

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 0
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.6089	1.0	29	0.1385	0.9962
0.1289	2.0	58	0.0510	0.9962
0.0835	3.0	87	0.0433	0.9885
0.0605	4.0	116	0.0330	0.9923
0.0479	5.0	145	0.0273	0.9904

Framework versions

• Transformers 4.31.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.13.1
• Tokenizers 0.13.3