whisper-tiny-speech-commands

This model is a fine-tuned version of openai/whisper-tiny on the speech_commands dataset. It achieves the following results on the evaluation set:

• Loss: 1.3232
• Accuracy: 0.8040

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 96
• eval_batch_size: 96
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.4229	1.0	412	1.1286	0.7936
0.1396	2.0	824	1.0506	0.7995
0.1323	3.0	1236	1.1224	0.7977
0.0528	4.0	1648	1.0810	0.8004
0.0889	5.0	2060	0.9224	0.8022
0.076	6.0	2472	1.0393	0.7981
0.0429	7.0	2884	1.1115	0.7990
0.0007	8.0	3296	1.1706	0.8026
0.0129	9.0	3708	1.0661	0.8013
0.0161	10.0	4120	1.0114	0.7990
0.0205	11.0	4532	1.2129	0.8031
0.0107	12.0	4944	1.1118	0.8026
0.0099	13.0	5356	0.9145	0.8031
0.0002	14.0	5768	1.1582	0.7999
0.0001	15.0	6180	1.2959	0.8035
0.0163	16.0	6592	1.0992	0.8026
0.0001	17.0	7004	1.2913	0.8035
0.0003	18.0	7416	1.3232	0.8040
0.0001	19.0	7828	1.3720	0.8040
0.0001	20.0	8240	1.3889	0.8040

Framework versions

• Transformers 4.43.3
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1