asr-africa/whisper_BIG-C_BEMBA_50hr_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: 1.2937
• Wer: 0.3800
• Cer: 0.1111

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: 4
• 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	Validation Loss	Wer	Cer
1.0801	1.0	6423	0.7376	0.5165	0.1630
0.615	2.0	12846	0.6440	0.4848	0.1654
0.4887	3.0	19269	0.6246	0.4440	0.1410
0.3615	4.0	25692	0.6347	0.4345	0.1403
0.2395	5.0	32115	0.6764	0.4336	0.1416
0.1389	6.0	38538	0.7458	0.4254	0.1354
0.0768	7.0	44961	0.8207	0.4357	0.1383
0.0473	8.0	51384	0.8766	0.4290	0.1361
0.0333	9.0	57807	0.9281	0.4209	0.1357
0.0266	10.0	64230	0.9782	0.4177	0.1351
0.0224	11.0	70653	1.0204	0.4299	0.1364
0.0191	12.0	77076	1.0563	0.4245	0.1381
0.0167	13.0	83499	1.0961	0.4182	0.1339
0.0179	14.0	89922	1.1083	0.4205	0.1381
0.0154	15.0	96345	1.1353	0.4236	0.1371
0.0131	16.0	102768	1.1734	0.4213	0.1364
0.0117	17.0	109191	1.2008	0.4158	0.1339
0.0104	18.0	115614	1.2104	0.4126	0.1347
0.0094	19.0	122037	1.2438	0.4084	0.1318
0.0089	20.0	128460	1.2541	0.4202	0.1343
0.0082	21.0	134883	1.2878	0.4124	0.1333
0.0078	22.0	141306	1.2901	0.4115	0.1339
0.0079	23.0	147729	1.3050	0.4100	0.1339
0.0067	24.0	154152	1.3187	0.4109	0.1351
0.0062	25.0	160575	1.3449	0.4107	0.1335
0.0065	26.0	166998	1.3667	0.4167	0.1344
0.0054	27.0	173421	1.3519	0.4134	0.1340
0.0054	28.0	179844	1.3755	0.4073	0.1329
0.0049	29.0	186267	1.3878	0.4107	0.1342
0.0052	30.0	192690	1.3963	0.4154	0.1326
0.0046	31.0	199113	1.4300	0.4106	0.1329
0.0042	32.0	205536	1.4242	0.4084	0.1330
0.0042	33.0	211959	1.4400	0.4066	0.1329
0.0039	34.0	218382	1.4541	0.4075	0.1323
0.0037	35.0	224805	1.4595	0.4090	0.1337
0.0035	36.0	231228	1.4757	0.4088	0.1314
0.0036	37.0	237651	1.4612	0.4065	0.1331
0.003	38.0	244074	1.4946	0.4081	0.1332

Framework versions

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