Project Beatrice V2 — Neural Voice Models

Official repository of pre-trained, high-fidelity neural voice models for the Beatrice real-time voice conversion engine.

---

📂 Repository Structure

Beatrice-voice-models/
├── assets/
│   └── beatrice-logo.png              # Official Project Beatrice V2 Logo
├── voice_models/
│   └── trump/                         # Trump Neural Voice Model (Step 5000)
│       ├── beatrice_paraphernalia_data_00005000.toml  # Model Manifest & Checkpoint Config
│       ├── waveform_generator.bin     # Neural Vocoder Waveform Generator (7.9 MB)
│       ├── phone_extractor.bin        # Phoneme Feature Extractor (7.2 MB)
│       ├── pitch_estimator.bin        # Pitch & Frequency Estimator (3.5 MB)
│       ├── embedding_setter.bin       # Speaker Embedding Setter (264 KB)
│       ├── speaker_embeddings.bin     # Target Speaker Vector Embeddings (234 KB)
│       ├── noimage.png                # Model Avatar Icon (388 KB)
│       └── README.md                  # Model-Specific Documentation
├── LICENSE                            # MIT Open-Source License
└── README.md                          # Repository Documentation & Usage Guide

---

🎙️ Available Voice Models (voice_models/)

1. Donald Trump Neural Voice (voice_models/trump/)

High-fidelity neural voice model trained on clean speech recordings, fully convergent at 5,000 steps for ultra-low latency real-time voice conversion.

File Name	Description	Size	Direct Download
beatrice_paraphernalia_data_00005000.toml	Model Configuration & Checkpoint Manifest	568 B	📥 Download Config
waveform_generator.bin	Neural Vocoder Waveform Generator Weights	7.9 MB	📥 Download Vocoder
phone_extractor.bin	Phoneme Feature Extractor Weights	7.2 MB	📥 Download Extractor
pitch_estimator.bin	Pitch & Frequency Estimator Weights	3.5 MB	📥 Download Pitch Model
embedding_setter.bin	Speaker Embedding Setter Weights	264 KB	📥 Download Embedding Setter
speaker_embeddings.bin	Target Speaker Vector Embeddings	234 KB	📥 Download Speaker Vector

---

⚡ How to Load Voice Models into Beatrice Voice Changer

Method 1: Clone the Entire Models Repository

git clone https://github.com/Project-Beatrice-V2/Beatrice-voice-models.git

Method 2: Load into Beatrice Voice Changer App

1. Open Beatrice Voice Changer on macOS (Metal/MPS) or Windows (CUDA).
2. Under Model / Target Voice, click Load Directory or select the manifest file:

   voice_models/trump/beatrice_paraphernalia_data_00005000.toml
   

3. Set your Microphone Input and Virtual Audio Output (BlackHole 2ch on macOS, VB-Audio Virtual Cable on Windows).
4. In Discord, OBS Studio, or game settings, select the virtual driver as your microphone to start streaming with transformed speech!

---

🛠️ Performance & Technical Specifications

• Latency: Glass-to-glass real-time inference latency <50ms.
• Audio Sample Rate: Native 48kHz high-fidelity output.
• Hardware Acceleration:
  • macOS: Apple Silicon Metal Performance Shaders (MPS).
  • Windows: NVIDIA CUDA & DirectX 12 DirectML.
• Model Package Size: ~19.5 MB total footprint per voice model.
• Hugging Face Mirror: SatiricalGuru/beatrice-voice-models

---

📜 License

This project is licensed under the MIT License — free for both personal and commercial use.

---

_{Maintained by the Project Beatrice V2 Open-Source Community.}