innovationm-ai/orpheus-tts-3b-finetuned

<div align="center"> <img src="./logo.png" alt="InnovationM"/> <br/><br/>

<a href="https://www.innovationm.com/"> <img src="https://img.shields.io/badge/InnovationM-Website-orange?style=for-the-badge&logo=google-chrome&logoColor=white" alt="InnovationM Website Badge"/> </a> <a href="https://in.linkedin.com/company/innovationm"> <img src="https://img.shields.io/badge/LinkedIn-Profile-blue?style=for-the-badge&logo=linkedin&logoColor=white" alt="LinkedIn Badge"/> </a> <br/><br/>

<a href="https://colab.research.google.com/drive/17ZeVLF8Q1P9juIKftijN9uwl2LrfMeCz?usp=drive_link"> 📓 Colab Notebook: Orpheus 3B TTS Finetuning </a> </div>

Orpheus-3B (Fine-Tuned) — InnovationM

This repository contains a fine-tuned variant of Orpheus TTS 3B — a state-of-the-art speech generation framework originally developed by Canopy Labs.

Fine-Tuned Model Description

This model is derived from the base checkpoint `canopylabs/orpheus-3b-0.1-ft`, fine-tuned using the Unsloth optimization framework, which enables up to 2× faster training with reduced memory footprint.

Additional alignment and adaptation were performed with the Hugging Face TRL (Training Reinforcement Learning) library.

Audio Samples

Before Finetuning (Female Voice)

<audio controls> <source src="https://huggingface.co/aiAtInnovationM/orpheus-tts-3b-finetuned/resolve/main/ttsbeforetuning_1.wav" type="audio/wav"> Your browser does not support the audio element. </audio>

After Finetuning (Male Voice)

<audio controls> <source src="https://huggingface.co/aiAtInnovationM/orpheus-tts-3b-finetuned/resolve/main/ttsaftertuning_1.wav" type="audio/wav"> Your browser does not support the audio element. </audio>

Fine-Tuned Model Information

• —Fine-tuned by: InnovationM's AI Team
• —Organization: InnovationM
• —License: Apache-2.0
• —Base Model: canopylabs/orpheus-3b-0.1-ft
• —Dataset Used: aiAtInnovationM/dailytalk-male

Special Thanks

• —Canopy Labs Research Team — for creating and open-sourcing the original CSM architecture and research.
• —Unsloth Team — for providing the high-efficiency training framework enabling optimized fine-tuning workflows.
• —Meta AI & Llama Team — for developing and releasing the foundational Llama models that support the Orpheus architecture and its derivatives.

Authors

Original Model Authors

Canopy Labs Team.

Fine-Tuned Model Authors

The InnovationM team.

📎 Useful Resources

📄 Citation

If you use this model in your research, please cite both Canopy Labs Model and this InnovationM fine-tuned version.

@misc{innovationm2025,
  title = {Orpheus-3B Fine-Tuned Model},
  author = {InnovationM},
  howpublished = {Hugging Face},
  year = {2025}
}