XLSR-WithLM-Malayalam

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the IMASC, Indic TTS Malayalam, OpenSLR Malayalam Train split datasets. It achieves the following results on the evaluation set:

• Loss: 0.1395
• Wer: 0.2952

Trigram Language Model Trained using KENLM Library on kavyamanohar/ml-sentences dataset

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.00024
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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: 800
• num_epochs: 1
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4912	0.1165	1000	0.5497	0.7011
0.5377	0.2330	2000	0.3292	0.5364
0.4343	0.3494	3000	0.2475	0.4424
0.3678	0.4659	4000	0.2145	0.4014
0.3345	0.5824	5000	0.1898	0.3774
0.3029	0.6989	6000	0.1718	0.3441
0.2685	0.8153	7000	0.1517	0.3135
0.2385	0.9318	8000	0.1395	0.2952

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1