asr-africa/whisper_NCHLT_speech_corpus_Xhosa_20hr_v1

Whisper Small Xhosa - Beijuka Bruno

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

• Loss: 2.5542
• Model Preparation Time: 0.0119
• Wer: 69.8439
• Cer: 41.1487

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	Model Preparation Time	Wer	Cer
0.4923	1.0	2344	0.2663	0.0119	25.4402	5.2566
0.1013	2.0	4688	0.2011	0.0119	17.9407	4.5978
0.0351	3.0	7032	0.1830	0.0119	15.7121	3.4549
0.0156	4.0	9376	0.1825	0.0119	14.3414	3.6433
0.0093	5.0	11720	0.1931	0.0119	14.6089	3.4396
0.0069	6.0	14064	0.1950	0.0119	13.6617	3.2572
0.0057	7.0	16408	0.1970	0.0119	13.7508	3.1117
0.0045	8.0	18752	0.1926	0.0119	13.2271	3.1684
0.0041	9.0	21096	0.1987	0.0119	13.0265	3.0489
0.0035	10.0	23440	0.2108	0.0119	12.8148	3.0228
0.0037	11.0	25784	0.1937	0.0119	12.5251	2.9416
0.0035	12.0	28128	0.2072	0.0119	12.9374	3.0152
0.003	13.0	30472	0.2066	0.0119	12.8594	2.9753
0.0031	14.0	32816	0.2045	0.0119	12.9374	3.0596
0.0028	15.0	35160	0.2023	0.0119	12.1351	2.9125
0.0027	16.0	37504	0.2007	0.0119	12.1685	2.9186
0.0019	17.0	39848	0.2091	0.0119	12.2799	2.9171
0.0022	18.0	42192	0.2106	0.0119	12.1908	2.9692
0.002	19.0	44536	0.2074	0.0119	11.9568	2.9232
0.0021	20.0	46880	0.2066	0.0119	12.6031	3.0887
0.0017	21.0	49224	0.2107	0.0119	12.0793	2.9355
0.0013	22.0	51568	0.2044	0.0119	11.6670	2.8083
0.0018	23.0	53912	0.2166	0.0119	12.2799	3.0152
0.0021	24.0	56256	0.2272	0.0119	12.0236	2.9186
0.0013	25.0	58600	0.2090	0.0119	12.0348	2.9937
0.0013	26.0	60944	0.2055	0.0119	11.5779	2.8022
0.0016	27.0	63288	0.2069	0.0119	11.7673	2.8742
0.0011	28.0	65632	0.2082	0.0119	11.4776	3.0504
0.0014	29.0	67976	0.2147	0.0119	11.7785	2.9983
0.0013	30.0	70320	0.2177	0.0119	12.4248	3.1071
0.001	31.0	72664	0.2136	0.0119	11.2993	2.9370
0.0012	32.0	75008	0.2136	0.0119	11.6336	2.9202
0.001	33.0	77352	0.2122	0.0119	11.2993	2.8328
0.0011	34.0	79696	0.2318	0.0119	11.3216	2.7808
0.0014	35.0	82040	0.2276	0.0119	11.6670	2.8466
0.0008	36.0	84384	0.2220	0.0119	11.4999	2.8359
0.0008	37.0	86728	0.2160	0.0119	11.3885	2.7486
0.0009	38.0	89072	0.2081	0.0119	10.9093	2.7486
0.001	39.0	91416	0.2085	0.0119	10.8870	2.7118
0.0005	40.0	93760	0.2142	0.0119	11.2547	2.8926
0.0006	41.0	96104	0.2153	0.0119	11.4776	2.9355
0.0007	42.0	98448	0.2107	0.0119	11.1656	2.8604
0.0009	43.0	100792	0.2152	0.0119	10.6641	2.6168
0.0007	44.0	103136	0.2134	0.0119	10.8870	2.7057
0.0006	45.0	105480	0.2154	0.0119	11.1656	2.8160
0.0007	46.0	107824	0.2199	0.0119	10.7979	2.7072
0.0004	47.0	110168	0.2255	0.0119	11.3105	2.8727
0.0004	48.0	112512	0.2229	0.0119	10.7087	2.8359
0.0003	49.0	114856	0.2190	0.0119	10.8424	2.7961
0.0005	50.0	117200	0.2182	0.0119	10.9762	2.8374
0.0006	51.0	119544	0.2172	0.0119	10.4301	2.6689
0.0007	52.0	121888	0.2080	0.0119	10.8536	3.3645
0.0009	53.0	124232	0.2162	0.0119	10.9762	2.8252
0.0004	54.0	126576	0.2157	0.0119	10.6641	2.7149
0.0001	55.0	128920	0.2126	0.0119	10.3410	2.6536
0.0003	56.0	131264	0.2150	0.0119	10.7421	2.7762
0.0003	57.0	133608	0.2166	0.0119	10.6307	2.6704
0.0002	58.0	135952	0.2211	0.0119	10.4413	2.6628
0.0004	59.0	138296	0.2178	0.0119	10.7533	2.8114
0.0002	60.0	140640	0.2241	0.0119	10.7533	2.7578
0.0002	61.0	142984	0.2280	0.0119	10.6753	2.6704
0.0	62.0	145328	0.2215	0.0119	10.3633	2.6092
0.0001	63.0	147672	0.2288	0.0119	10.8759	2.7547
0.0003	64.0	150016	0.2267	0.0119	10.8424	2.7072
0.0001	65.0	152360	0.2228	0.0119	10.5416	2.6827

Framework versions

• Transformers 4.47.1
• Pytorch 2.1.0+cu118
• Datasets 3.2.0
• Tokenizers 0.21.0