voicestudio 1.0.0

VoiceStudio: A unified toolkit for text-style prompted speech synthesis, voice adaptation, and editing

VoiceStudio

Your Complete Voice Adaptation Workspace

Installation | Quick Start | Documentation | Papers

---

🎯 Overview

VoiceStudio is a unified toolkit for text-style prompted speech synthesis, enabling instant voice adaptation and editing through natural language descriptions. Built on cutting-edge research in voice style prompting, LoRA adaptation, and language-audio models.

Key Features:

• 🎨 Text-Style Prompting: Control voice characteristics with natural language
• ⚡ Instant Adaptation: Real-time LoRA generation for any TTS model
• ✂️ Voice Editing: Modify existing voices with simple instructions
• 🔧 Architecture Agnostic: Works with multiple TTS architectures
• 🚀 Production Ready: Optimized for both research and deployment

---

🆕 What's New

v0.1.0 (2025)

• 🔍 Speaker consistency analysis tools
• 🎨 BOS token P-tuning
• 📊 Attention visualization

---

🚀 Installation

From PyPI (Recommended)

uv add voicestudio[all]

From Source

uv add git+https://github.com/LatentForge/voicestudio.git

Requirements

• Python 3.8+
• PyTorch 2.0+
• CUDA 11.8+ (for GPU acceleration)

---

📚 Advanced Usage

Custom TTS Model Integration

VoiceStudio supports any TTS model through a simple adapter interface:

from voicestudio import TTSAdapter, LoRAGenerator

# Wrap your TTS model
class MyTTSAdapter(TTSAdapter):
    def __init__(self, model):
        self.model = model
    
    def get_lora_target_modules(self):
        return ["attention.q_proj", "attention.v_proj"]
    
    def forward(self, text, lora_weights=None):
        if lora_weights:
            self.apply_lora(lora_weights)
        return self.model(text)

# Use with VoiceStudio
adapter = MyTTSAdapter(my_tts_model)
generator = LoRAGenerator.from_pretrained("voicestudio/t2a-lora-base")

lora = generator("professional news anchor voice")
audio = adapter(text="Breaking news tonight...", lora_weights=lora)

Multi-Speaker Voice Blending

from voicestudio import VoiceBlender

blender = VoiceBlender()

# Blend multiple voice characteristics
blended_lora = blender.blend([
    ("warm and friendly", 0.6),
    ("professional and clear", 0.4)
])

audio = tts_model.synthesize(text, lora=blended_lora)

Fine-tuning on Custom Data

from voicestudio import LoRAGenerator
from voicestudio.training import Trainer

# Load pre-trained generator
generator = LoRAGenerator.from_pretrained("voicestudio/t2a-lora-base")

# Fine-tune on your data
trainer = Trainer(
    model=generator,
    train_dataset=your_dataset,
    output_dir="./checkpoints"
)

trainer.train()

---

📊 Supported Models

VoiceStudio works with various TTS architectures:

Model	Status	Notes
VITS	✅ Supported	Fully tested
FastSpeech2	✅ Supported	Fully tested
Tacotron2	✅ Supported	Requires adapter
VALL-E	🔄 Experimental	Work in progress
Bark	🔄 Experimental	Coming soon
YourTTS	✅ Supported	Community contributed

Add your own model: See our Integration Guide

---

@inproceedings{voicestudio2027lam,
  title={T2A-LoRA2: Text-Guided Voice Editing with Language-Audio Models},
  author={Your Name},
  booktitle={ICML},
  year={2027}
}

---

---

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Areas we need help with:

• 🔧 Additional TTS model adapters
• 📚 Documentation improvements
• 🐛 Bug fixes and testing
• 🌍 Multi-language support
• 🎨 New voice editing techniques

---

📝 License

This project is licensed under the MIT License - see LICENSE file for details.

---

🙏 Acknowledgments

• CLAP: Microsoft & LAION-AI for CLAP model
• LoRA: Microsoft for LoRA technique
• HuggingFace: For transformers library and model hub
• LatentForge Team: For research support and infrastructure

---

🌟 Citation

If you use VoiceStudio in your research, please cite:

@software{voicestudio2026,
  title={VoiceStudio: A Unified Toolkit for Voice Style Adaptation},
  author={Your Name},
  year={2026},
  url={https://github.com/LatentForge/voicestudio}
}

---

Made with ❤️ by the LatentForge Team

⭐ Star us on GitHub | 📖 Read the Docs | 🤗 HuggingFace