wav2vec2_transformer_phonome

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2714
• Wer: 0.5886
• Cer: 0.0707

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: 3e-05
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2000
• training_steps: 26000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.0354	1.49	1000	1.5671	0.9984	0.5492
1.3073	2.98	2000	0.5049	0.7604	0.1035
1.1054	4.46	3000	0.3268	0.6848	0.0865
1.066	5.95	4000	0.3185	0.6734	0.0814
1.0249	7.44	5000	0.3240	0.6483	0.0796
0.9736	8.93	6000	0.3017	0.6206	0.0778
0.9367	10.42	7000	0.2813	0.6279	0.0752
0.8958	11.9	8000	0.2778	0.6117	0.0763
0.8778	13.39	9000	0.2772	0.6393	0.0765
0.9016	14.88	10000	0.2768	0.6271	0.0751
0.8208	16.37	11000	0.3309	0.6182	0.0759
0.8297	17.86	12000	0.2814	0.6011	0.0721
0.7533	19.35	13000	0.2674	0.6068	0.0733
0.7959	20.83	14000	0.2821	0.6206	0.0736
0.7577	22.32	15000	0.3250	0.6206	0.0735
0.7456	23.81	16000	0.3078	0.5979	0.0742
0.7387	25.3	17000	0.3166	0.5930	0.0720
0.7364	26.79	18000	0.3052	0.6141	0.0739
0.7136	28.27	19000	0.3026	0.6060	0.0731
0.7036	29.76	20000	0.2726	0.5946	0.0720
0.6939	31.25	21000	0.2714	0.5930	0.0720
0.6985	32.74	22000	0.2722	0.5963	0.0713
0.677	34.23	23000	0.2799	0.6011	0.0718
0.7176	35.71	24000	0.2769	0.6052	0.0710
0.6649	37.2	25000	0.2751	0.5987	0.0716
0.642	38.69	26000	0.2719	0.5963	0.0704

Framework versions

• Transformers 4.17.0
• Pytorch 2.4.0
• Datasets 3.0.1
• Tokenizers 0.20.0