---
language:
- en
license: mit
library_name: gguf
pipeline_tag: text-generation
tags:
- gguf
- conversational
- text-generation
- text-to-speech
- voice-cloning
- offline
- digital-twin
- xtts
- piper-tts
- whisper
---

# Jason AI Twin (Offline Digital Twin & Cloned Voice)

An offline, privacy-first AI Digital Twin architecture featuring quantized local LLM reasoning, custom knowledge bases, GPU-accelerated speech recognition, and neural voice cloning.

## 🎙️ Cloned Voice & Audio Assets

The AI Twin features neural voice cloning sourced from `/local-tts`:

* **`cloned_output.wav`**: High-fidelity cloned voice reference sample synthesized from Jason's voice profile. Used as the primary speaker reference for zero-shot Coqui XTTS v2 voice synthesis.
* **`cloned_sample2.wav`**: Second verified neural synthesis sample generated with Coqui XTTS v2 on CUDA using Jason's vocal profile ("The voice cloning matrix is fully operational...").
* **`my_voice_clean.wav`**: Cleaned, noise-reduced studio vocal sample of Jason's voice (22.05 kHz PCM) for fine-tuning and voice cloning alignment.
* **`amy.onnx` / `amy.onnx.json`**: Ultra-fast, low-latency ONNX voice model for Piper TTS fallback.

---

## 🧠 Brain & Models

* **`digital_twin_q4.gguf`**: Q4_K_M quantized LLM brain (`jason_twin`) trained and adapted on Jason's communication style, knowledge bases, and preferences.
* **`Modelfile`**: Ollama configuration for launching `jason_twin` locally with optimized system prompt, temperature, and context length.

---

## 🚀 Interactive CLI Client (`chat_twin.py`)

The repository includes [chat_twin.py](chat_twin.py), an interactive voice and text terminal client:
* **Two-Way Voice Interaction**: Press `[ENTER]` on an empty prompt to record your voice; local Whisper (`openai/whisper-tiny.en`) transcribes it with GPU acceleration.
* **Neural Voice Synthesis**: Speaks responses back in Jason's cloned voice using local XTTS v2 (`--speaker_wav cloned_output.wav`).
* **Multi-Engine Fallback**: Gracefully falls back to Piper ONNX, `spd-say`, or `espeak-ng` if GPU voice cloning is unavailable.

### In-Chat Commands:
| Command | Action |
| :--- | :--- |
| `/voice cloned` | Activate Jason's cloned voice from `/local-tts` |
| `/voice <type>` | Switch to system voices (`male1`, `male2`, `female1`, etc.) |
| `/rate <val>` | Adjust speech rate (-100 to 100) |
| `/tts on\|off` | Toggle Text-to-Speech output |
| `/stop` or `/s` | Immediately silence current speech output |
| `/clear` or `/c` | Reset conversation context and clear terminal |
| `/help` or `/h` | Display digital twin command menu |
| `/exit` | Exit the application |

---

## 🛠️ Quickstart

### 1. Run via Ollama
```bash
ollama create jason_twin -f Modelfile
ollama run jason_twin
```

### 2. Run Interactive Voice Twin CLI
```bash
# Ensure dependencies are installed
pip install TTS transformers torch torchaudio

# Launch voice chat client
python3 chat_twin.py
```

### 3. Desktop Shortcut
On Ubuntu systems, launch directly via the included `AI_Twin.desktop` or:
```bash
./launch_twin.sh
```