Whisper Base Shona + Tshivenda Pilot

This model is a fine-tuned version of openai/whisper-base on the Cleaned WAXAL Shona and the Swivuriso Tshivenda datasets. It achieves the following results on the evaluation set:

• eval_loss: 0.6685
• eval_wer: 56.4808
• eval_shona_wer: 36.8120
• eval_shona_loss: 0.4379
• eval_tshivenda_wer: 76.1495
• eval_tshivenda_loss: 0.8991
• epoch: 8.8183
• step: 20000

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 20000
• mixed_precision_training: Native AMP

Framework versions

• Transformers 5.15.0
• Pytorch 2.13.0+cu130
• Datasets 5.0.1
• Tokenizers 0.22.2