This model is a fine-tuned version of openai/whisper-small on the IWSLT-2023, FLEURS, BiteSize, and SpokenWords dataset. It achieves the following results on the evaluation set:

Loss: 1.8231
Bleu: 29.51
Chrf: 44.29
Wer: 67.0869

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: 0.0001
train_batch_size: 64
eval_batch_size: 64
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
training_steps: 4000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Bleu	Chrf	Validation Loss	Wer
1.9416	0.2155	100	13.09	26.48	1.7899	104.4575
1.5186	0.4310	200	18.6	35.75	1.5696	87.5732
1.2884	0.6466	300	17.57	37.19	1.4751	87.2580
1.0729	0.8621	400	17.92	38.23	1.4345	99.2346
0.4574	1.0776	500	22.48	39.17	1.5585	83.1607
0.4517	1.2931	600	22.53	38.38	1.5763	81.7650
0.4385	1.5086	700	20.05	39.46	1.5852	96.8483
0.3934	1.7241	800	26.89	42.67	1.5332	70.6889
0.3587	1.9397	900	28.95	44.16	1.5025	64.9707
0.1528	2.1552	1000	28.32	42.36	1.5882	65.8712
0.1425	2.3707	1100	25.5	42.42	1.6056	75.0113
0.1389	2.5862	1200	26.52	42.11	1.6236	70.6439
0.1532	2.8017	1300	25.78	41.61	1.6196	75.9118
0.1138	3.0172	1400	26.01	40.88	1.7185	69.6983
0.0661	3.2328	1500	28.74	43.16	1.6626	71.2292
0.0625	3.4483	1600	29.16	43.6	1.6835	66.3215
0.0615	3.6638	1700	28.93	44.08	1.6756	68.3476
0.0611	3.8793	1800	27.77	43.67	1.6648	72.1747
0.0344	4.0948	1900	28.33	44.18	1.7351	68.1225
0.0339	4.3103	2000	28.9	42.98	1.7715	67.0869
0.0369	4.5259	2100	29.83	44.87	1.7200	64.8807
0.0326	4.7414	2200	28.23	43.75	1.7232	69.3832
0.0346	4.9569	2300	27.72	43.1	1.7688	72.8050
0.0167	5.1724	2400	28.73	43.26	1.8072	67.4471
0.0146	5.3879	2500	29.91	44.24	1.7801	66.4566
0.0165	5.6034	2600	29.34	44.33	1.7782	68.2125
0.0143	5.8190	2700	27.78	43.07	1.7675	72.5799
0.0106	6.0345	2800	29.45	43.31	1.7660	67.5371
0.0098	6.25	2900	27.89	42.67	1.7803	71.6344
0.0087	6.4655	3000	27.66	43.04	1.7786	72.0396
0.0089	6.6810	3100	1.7661	29.81	44.65	67.3120
0.0081	6.8966	3200	1.7744	29.48	44.3	68.0324
0.0095	7.1121	3300	1.8197	29.55	44.2	67.5371
0.0112	7.3276	3400	1.8102	29.34	43.9	66.2765
0.0075	7.5431	3500	1.8004	29.57	44.43	67.3570
0.0111	7.7586	3600	1.8015	29.56	44.57	66.4566
0.009	7.9741	3700	1.8001	29.7	45.24	66.6817
0.005	8.1897	3800	1.8184	29.21	44.4	67.4471
0.0055	8.4052	3900	1.8222	29.67	44.35	67.1319
0.0042	8.6207	4000	1.8231	29.51	44.29	67.0869

Framework versions

Transformers 4.40.2
Pytorch 2.2.0+cu121
Datasets 2.19.1
Tokenizers 0.19.1

ymoslem/whisper-small-ga2en-v1.2.1-r

Whisper Small GA-EN Speech Translation

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

ymoslem/
whisper-small-ga2en-v1.2.1-r