Bardic AI

Local AI-powered D&D DM with voice input/output. Ollama + Whisper + Coqui TTS for offline tabletop adventures.

Features

• Voice Input: Speak your actions using Whisper speech-to-text
• Voice Output: DM responses with voice cloning using Coqui TTS XTTS v2
• Offline: Runs completely locally on your machine
• Customizable Voices: Clone your own voice or create unique NPC voices
• GPU Accelerated: Uses CUDA for fast TTS and Whisper transcription

Setup

Prerequisites

• Python 3.9+
• CUDA-capable GPU (12GB+ VRAM recommended)
• Ollama installed and running
• Qwen 2.5 14B model: ollama pull qwen2.5:14b

Installation

# Clone the repo
git clone https://github.com/yourusername/bardic-ai.git
cd bardic-ai

# Create virtual environment
python -m venv venv

# Activate virtual environment
# On Linux/Mac:
source venv/bin/activate
# On Windows:
venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Copy environment file
cp .env.example .env

Recording Your Voice Sample

Before running the app, record a voice sample for voice cloning:

python record_voice_sample.py

This will:

1. Record 30 seconds of audio from your microphone
2. Show a countdown timer while recording
3. Save to voice_samples/dm_narrator.wav

Tips for best results:

• Speak naturally and clearly
• Vary your tone and intonation
• Record in a quiet room with minimal background noise
• Position microphone 6-8 inches from your mouth

See voice_samples/README.md for detailed voice recording guidelines.

Running the App

python app.py

Visit http://localhost:5000 to start your adventure!

• Use text input for traditional gameplay
• Click "Use Voice Input" for voice-powered gameplay

Voice Cloning

Speed Adjustment

Adjust speech speed in services/voice_output.py:

text_to_speech(text, voice_sample="voice_samples/dm_narrator.wav", speed=1.2)

• speed=0.5 - Very slow (dramatic moments)
• speed=1.0 - Normal speaking pace
• speed=1.2 - Slightly faster (default, keeps gameplay moving)
• speed=1.5 - Fast narration

Multiple Voice Samples for NPCs

Create different voice samples for different characters:

# Record and rename for specific NPCs
python record_voice_sample.py
mv voice_samples/dm_narrator.wav voice_samples/npc_merchant.wav

python record_voice_sample.py
mv voice_samples/dm_narrator.wav voice_samples/npc_villain.wav

Then update your code to use different voices for different NPCs by passing the voice_sample parameter.

Technical Details

• LLM: Ollama (Qwen 2.5 14B)
• Speech-to-Text: OpenAI Whisper (base model)
• Text-to-Speech: Coqui TTS XTTS v2 with voice cloning
• Web Framework: Flask
• Audio Processing: sounddevice, scipy, numpy

Roadmap

• Basic Ollama integration
• Voice input (Whisper)
• Voice output (Coqui TTS with voice cloning)
• Game state persistence
• Combat system
• Campaign structure
• Multi-voice NPC support
• Real-time voice conversation mode