RobCaamano/Sherlock-Holmes

This model is a fine-tuned version of XTTS-v2 on a custom dataset.

Training and evaluation data

Audio collected from audiobook.

Training hyperparameters

The following hyperparameters were used during training:

GPTArgs():

• max_conditioning_length=143677
• min_conditioning_length=66150
• max_wav_length=255995
• max_text_length=66150
• gpt_use_masking_gt_prompt_approach=True
• gpt_use_perceiver_resampler=True

GPTTrainerConfig():

• BATCH_SIZE=3
• batch_group_size=48
• GRAD_ACUMM_STEPS=84
• optimizer_params={"betas": [0.9, 0.96], "eps": 1e-8, "weight_decay": 1e-2}
• lr_scheduler_params={"milestones": [50000 * 18, 150000 * 18, 300000 * 18], "gamma": 0.5, "last_epoch": -1}

Training results

• Train avg_loss: 0.05243
• Train avg_loss_mel_ce: 4.38085
• Train avg_loss_text_ce: 0.02336
• Validation avg_loss: 4.1927
• Validation avg_loss_mel_ce: 4.17117
• Validation avg_loss_text_ce: 0.02153
• Epoch: 1

Framework versions

• Transformers 4.51.3
• PyTorch 2.6.0+cu126
• TorchAudio 2.6.0
• Tokenizers 0.21.1
• TTS