asr-africa/whisper_DigitalUmuganda_Afrivoice_Shona_5hr_v1

Whisper Small Shona - Beijuka Bruno

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

• Loss: 1.0076
• Wer: 48.3697
• Cer: 11.1051

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.9327	1.0	110	1.2601	74.2137	18.0635
0.96	2.0	220	0.7862	54.7179	11.5259
0.5827	3.0	330	0.6185	47.9531	11.3006
0.3706	4.0	440	0.5590	40.7639	8.3464
0.2221	5.0	550	0.5628	40.1897	8.2400
0.1173	6.0	660	0.5671	42.4114	10.2366
0.06	7.0	770	0.5760	40.6141	10.4056
0.033	8.0	880	0.6091	37.2941	7.6798
0.0194	9.0	990	0.6050	36.9945	7.6016
0.0127	10.0	1100	0.6379	36.8198	7.7549
0.0081	11.0	1210	0.6320	36.7449	7.5014
0.0069	12.0	1320	0.6457	36.7948	7.3950
0.0048	13.0	1430	0.6605	37.1193	7.5484
0.0043	14.0	1540	0.6602	36.5701	7.7925
0.0027	15.0	1650	0.6601	36.1957	7.6391
0.0019	16.0	1760	0.6659	35.3220	7.1321
0.0017	17.0	1870	0.6741	35.8213	7.2667
0.0023	18.0	1980	0.6773	36.6201	8.1492
0.0018	19.0	2090	0.6804	35.5966	7.1759
0.0037	20.0	2200	0.6929	35.5467	7.3011
0.0035	21.0	2310	0.6816	36.1208	7.3762
0.0028	22.0	2420	0.6940	35.5716	7.2198
0.0017	23.0	2530	0.6873	35.9710	7.2980
0.0023	24.0	2640	0.7152	36.1208	7.3387
0.004	25.0	2750	0.6968	35.7963	7.3575
0.0032	26.0	2860	0.7012	37.3190	7.5984

Framework versions

• Transformers 4.44.2
• Pytorch 2.1.0+cu118
• Datasets 3.0.0
• Tokenizers 0.19.1