NbAiLab/whisper-large-sme

Whisper Large Northern Sámi

This model is a fine-tuned version of openai/whisper-large-v2 on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.5559
• Wer: 24.9143

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: 12
• eval_batch_size: 6
• seed: 42
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 60000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4665	58.0	1000	0.8572	54.5143
0.3041	117.0	2000	0.6711	44.1143
0.2671	176.0	3000	0.5794	39.7714
0.1761	235.0	4000	0.5357	35.0857
0.2089	294.0	5000	0.5094	33.6
0.1456	352.0	6000	0.4959	33.0286
0.1514	411.0	7000	0.4864	32.5714
0.1203	470.0	8000	0.4625	31.4286
0.0879	529.0	9000	0.4916	45.4857
0.0825	588.0	10000	0.4962	30.6286
0.0753	647.0	11000	0.4723	31.2
0.0812	705.0	12000	0.4574	28.6857
0.062	764.0	13000	0.4628	28.8000
0.0604	823.0	14000	0.4668	28.0000
0.0666	882.0	15000	0.4697	28.6857
0.0405	941.0	16000	0.4908	54.6286
0.0349	999.0	17000	0.4728	28.4571
0.0409	1058.0	18000	0.4884	28.4571
0.0292	1117.0	19000	0.4576	27.3143
0.0247	1176.0	20000	0.4734	28.9143
0.0229	1235.0	21000	0.4899	29.9429
0.0271	1294.0	22000	0.4790	28.1143
0.0271	1352.0	23000	0.5012	30.1714
0.0184	1411.0	24000	0.5008	27.3143
0.0211	1470.0	25000	0.5118	27.6571
0.0183	1529.0	26000	0.5398	30.0571
0.0164	1588.0	27000	0.5006	27.3143
0.0169	1647.0	28000	0.5059	27.0857
0.0147	1705.0	29000	0.5325	27.7714
0.0104	1764.0	30000	0.4818	26.1714
0.0128	1823.0	31000	0.5259	28.3429
0.0145	1882.0	32000	0.5299	26.2857
0.0075	1941.0	33000	0.5082	27.4286
0.0087	1999.0	34000	0.5144	26.6286
0.005	2058.0	35000	0.5590	27.0857
0.0099	2117.0	36000	0.5546	28.9143
0.007	2176.0	37000	0.5364	26.8571
0.0045	2235.0	38000	0.5574	27.2000
0.0064	2294.0	39000	0.5051	25.7143
0.0079	2352.0	40000	0.5247	25.9429
0.0083	2411.0	41000	0.5514	25.6
0.0101	2470.0	42000	0.5710	25.6
0.0062	2529.0	43000	0.5830	28.0000
0.0046	2588.0	44000	0.5828	26.8571
0.0053	2647.0	45000	0.5621	27.4286
0.0047	2705.0	46000	0.5673	25.9429
0.0045	2764.0	47000	0.5220	25.6
0.0065	2823.0	48000	0.5704	27.7714
0.0039	2882.0	49000	0.5741	27.7714
0.0027	2941.0	50000	0.5762	26.0571
0.0019	2999.0	51000	0.5559	24.9143
0.0015	3058.0	52000	0.5777	28.5714
0.0026	3117.0	53000	0.5589	25.2571
0.0032	3176.0	54000	0.6061	26.9714
0.0025	3235.0	55000	0.5776	25.1429
0.0046	3294.0	56000	0.5753	27.3143
0.0015	3352.0	57000	0.5736	27.2000
0.003	3411.0	58000	0.5933	25.6
0.002	3470.0	59000	0.6036	25.6
0.0007	58.0	60000	0.5975	25.2571

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.11.0