asr-300m-turkish

This model is a fine-tuned version of wav2vec2 xls-r 300M on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.4060
• Wer: 0.3879

Model description

The following datasets were used for finetuning:

• Common Voice 7.0 TR All validated split except test split was used for training.
• MediaSpeech

Intended uses & limitations

More information needed

Training and evaluation data

The following datasets were used for finetuning:

Common Voice 7.0 TR All validated split except test split was used for training. MediaSpeech

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.9649	4.26	400	0.7179	0.7544
0.3912	8.51	800	0.4798	0.5432
0.1848	12.77	1200	0.4588	0.4792
0.1277	17.02	1600	0.4676	0.4510
0.0923	21.28	2000	0.4251	0.4218
0.07	25.53	2400	0.4164	0.4006
0.0546	29.79	2800	0.4060	0.3879

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.1+cu118
• Datasets 1.18.3
• Tokenizers 0.13.3