asr-africa/whisper_NCHLT_speech_corpus_Zulu_50hr_v1

Whisper Small Shona - Beijuka Bruno

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

• Loss: 1.3971
• Wer: 58.9779
• Cer: 19.2528

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: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3596	1.0	5045	0.3025	28.6289	4.3976
0.1187	2.0	10090	0.2591	24.1626	3.7419
0.0531	3.0	15135	0.2731	22.8227	3.6080
0.0239	4.0	20180	0.2875	21.5721	3.6466
0.0129	5.0	25225	0.3149	21.9071	3.4945
0.0087	6.0	30270	0.3239	20.5449	3.2903
0.007	7.0	35315	0.3394	20.7459	3.3448
0.0059	8.0	40360	0.3542	19.9866	3.1496
0.0052	9.0	45405	0.3570	20.9692	3.3879
0.0044	10.0	50450	0.3667	20.3886	3.3107
0.0046	11.0	55495	0.3692	20.2099	3.3130
0.0039	12.0	60540	0.3664	20.8352	3.9007
0.0044	13.0	65585	0.3874	20.3662	3.6261
0.0038	14.0	70630	0.3915	20.5672	3.3561
0.0032	15.0	75675	0.3842	20.8352	3.4083
0.0029	16.0	80720	0.4151	20.7459	3.4128
0.0029	17.0	85765	0.4071	20.6789	3.3039
0.0026	18.0	90810	0.4144	20.1653	3.2767

Framework versions

• Transformers 4.47.0
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.21.0