MSP-ASR

This model is a fine-tuned version of facebook/wav2vec2-large-robust-ft-libri-960h on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3481
• Wer: 0.2040

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: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 1000.0
• training_steps: 40000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.9991	0.025	1000	0.7600	0.4273
2.7153	0.05	2000	0.5967	0.3697
2.4461	0.075	3000	0.5058	0.3102
2.4391	0.1	4000	0.5428	0.2994
2.4508	0.125	5000	0.6064	0.3583
2.0697	0.15	6000	0.4450	0.2544
2.0818	0.175	7000	0.7904	0.3479
2.1906	0.2	8000	0.6347	0.3174
2.1315	0.225	9000	0.5940	0.3114
2.0303	0.25	10000	0.7855	0.3289
1.9732	0.275	11000	0.4407	0.2292
1.7688	0.3	12000	0.9236	0.3608
2.0256	0.325	13000	0.4565	0.2409
2.1277	0.35	14000	0.6548	0.3096
1.9222	0.375	15000	0.4132	0.2314
1.8986	0.4	16000	0.3661	0.2074
1.9326	0.425	17000	0.3481	0.2040
1.9936	0.45	18000	0.5246	0.2579
1.9033	0.475	19000	0.4698	0.2397
1.8331	0.5	20000	0.4469	0.2189
2.0719	0.525	21000	0.6117	0.2701
1.8486	0.55	22000	0.4878	0.2329
1.7071	0.575	23000	0.6653	0.2782
1.7644	0.6	24000	0.6700	0.2846
1.6879	0.625	25000	0.7342	0.2891
1.9840	0.65	26000	0.8277	0.3100
1.7513	0.675	27000	0.6867	0.2832
1.7917	0.7	28000	0.5828	0.2630
1.9621	0.725	29000	0.4499	0.2243
1.8372	0.75	30000	0.5036	0.2397
1.8334	0.775	31000	0.5540	0.2513
1.7985	0.8	32000	0.6490	0.2780
1.7205	0.825	33000	0.5828	0.2615
2.1699	0.85	34000	0.6067	0.2732
1.7843	0.875	35000	0.5375	0.2540
1.8201	0.9	36000	0.5541	0.2564
1.6863	0.925	37000	0.5362	0.2512
1.7110	0.95	38000	0.5339	0.2495
1.7379	0.975	39000	0.5395	0.2505
1.8971	1.0	40000	0.5416	0.2509

Framework versions

• Transformers 5.10.2
• Pytorch 2.10.0+rocm7.2.4.git3d3aa833
• Datasets 4.0.0
• Tokenizers 0.22.2