asr-africa/whisper_BIG-C_BEMBA_100hr_v1

Whisper Small Bemba - Beijuka Bruno

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

• Loss: 1.3722
• Wer: 0.3520
• Cer: 0.1020

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.025
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
0.9576	1.0	12851	0.1144	0.5736	0.4270
0.5631	2.0	25702	0.1223	0.4973	0.3961
0.4616	3.0	38553	0.0990	0.4716	0.3559
0.3579	4.0	51404	0.1020	0.4812	0.3573
0.2534	5.0	64255	0.0969	0.5052	0.3472
0.1588	6.0	77106	0.0969	0.5568	0.3525
0.0922	7.0	89957	0.1019	0.6221	0.3587
0.0567	8.0	102808	0.0984	0.6833	0.3532
0.0407	9.0	115659	0.0981	0.7381	0.3509
0.0324	10.0	128510	0.7663	0.3504	0.0989
0.027	11.0	141361	0.8045	0.3505	0.0992
0.0235	12.0	154212	0.8415	0.3456	0.0974
0.0207	13.0	167063	0.8597	0.3445	0.0973
0.0184	14.0	179914	0.8730	0.3490	0.0986
0.0165	15.0	192765	0.9096	0.3439	0.0980
0.0154	16.0	205616	0.9381	0.3458	0.0979
0.0137	17.0	218467	0.9630	0.3468	0.0959
0.0129	18.0	231318	0.9761	0.3434	0.0983
0.0117	19.0	244169	0.9945	0.3434	0.0976
0.0109	20.0	257020	1.0103	0.3437	0.0967
0.0099	21.0	269871	1.0291	0.3466	0.0969
0.0097	22.0	282722	1.0475	0.3439	0.0976
0.0089	23.0	295573	1.0743	0.3380	0.0950
0.0084	24.0	308424	1.0840	0.3387	0.0958
0.0075	25.0	321275	1.1084	0.3396	0.0966
0.0074	26.0	334126	1.1091	0.3397	0.0986
0.0069	27.0	346977	1.1218	0.3385	0.0972
0.0063	28.0	359828	1.1461	0.3386	0.0963
0.0062	29.0	372679	1.1644	0.3402	0.0960
0.0058	30.0	385530	1.1612	0.3365	0.0952
0.0055	31.0	398381	1.1764	0.3354	0.0953
0.0052	32.0	411232	1.1749	0.3352	0.0957
0.0051	33.0	424083	1.1910	0.3399	0.0976
0.0046	34.0	436934	1.1948	0.3357	0.0958
0.0044	35.0	449785	1.2069	0.3359	0.0955
0.0043	36.0	462636	1.2228	0.3377	0.0957
0.004	37.0	475487	1.2419	0.3333	0.0952
0.0038	38.0	488338	1.2410	0.3354	0.0960
0.0036	39.0	501189	1.2430	0.3356	0.0952
0.0034	40.0	514040	1.2685	0.3358	0.0957
0.0033	41.0	526891	1.2591	0.3354	0.0962
0.003	42.0	539742	1.2770	0.3362	0.0952
0.003	43.0	552593	1.2896	0.3327	0.0950
0.0028	44.0	565444	1.2898	0.3314	0.0945
0.0026	45.0	578295	1.3017	0.3322	0.0946
0.0025	46.0	591146	1.3097	0.3307	0.0940
0.0024	47.0	603997	1.3177	0.3322	0.0941
0.0023	48.0	616848	1.3218	0.3285	0.0933
0.0021	49.0	629699	1.3259	0.3323	0.0945
0.0022	50.0	642550	1.3539	0.3301	0.0931
0.0019	51.0	655401	1.3442	0.3291	0.0941
0.0018	52.0	668252	1.3369	0.3324	0.0950
0.0018	53.0	681103	1.3489	0.3305	0.0941
0.0017	54.0	693954	1.3617	0.3294	0.0932
0.0015	55.0	706805	1.3495	0.3319	0.0946
0.0014	56.0	719656	1.3689	0.3311	0.0952
0.0013	57.0	732507	1.3870	0.3302	0.0933
0.0013	58.0	745358	1.3848	0.3289	0.0928

Framework versions

• Transformers 4.45.0
• Pytorch 2.1.0+cu118
• Datasets 3.0.0
• Tokenizers 0.20.0