Voice Training Data Creator

A desktop GUI application for creating voice training datasets with AI-powered synthetic text generation. Supports voice cloning, TTS model training, and STT fine-tuning workflows.

Screenshots

Dataset System

Each sample is saved as a numbered folder (starting from 001) with:

• Recording (as .wav)
• Text as .txt, incorporating user edits to scrub AI messages from source of truth
• JSON for metadata

Features

Audio Recording

• High-quality WAV recording (44.1kHz/48kHz, 16-bit)
• Preferred microphone selection (auto-selected on recording)
• Microphone selection and testing
• Real-time duration tracking
• Pause/resume functionality during recording
• Retake button: One-click to discard and restart recording
• Delete recording option

Text Generation

• Synthetic text generation using OpenAI GPT models
• Multiple style options:
  • General Purpose
  • Colloquial
  • Voice Note
  • Technical
  • Prose
• Custom vocabulary dictionary support
• Configurable duration and speaking rate (WPM)
• Post-generation text editing
• Regenerate option (keep audio, new text)
• New Sample button: Clear state and start completely fresh
• Narration view with adjustable font size

Sample Management

• Organized directory structure (numbered folders)
• Automatic sample numbering
• Metadata tracking (generation parameters, timestamps)
• Session statistics
• Total duration estimation
• Personalized goal tracking: Set target duration and track progress
• Quick access to dataset folder

Configuration and Settings

• Dedicated Settings Tab for easy access
• Secure API key storage (system keyring)
• API connection testing
• Preferred microphone configuration
• Configurable audio quality settings (sample rate)
• Multiple OpenAI model support (gpt-4o-mini, gpt-4o, gpt-4.1-mini)
• Persistent base path configuration
• Training goal duration (personalizes progress tracking)
• Auto-generate next sample preference

User Experience

• Three-tab interface (Record & Generate, Dataset, Settings)
• Light/dark theme toggle
• Helpful tooltips and descriptions
• Real-time statistics updates
• Clear workflow with New Sample and Retake buttons

Installation

Option 1: Install .deb Package (Recommended)

Standard installation method for Ubuntu/Debian systems:

1. Download or build the package:

   # Option A: Build from source
   git clone https://github.com/danielrosehill/Voice-Training-Data-Creator.git
   cd Voice-Training-Data-Creator
   ./build-deb.sh
   
   # Option B: Download release (when available)
   # Download voice-training-data-creator_1.0.0_all.deb from releases

2. Install the package:

   sudo dpkg -i voice-training-data-creator_1.0.0_all.deb
   sudo apt-get install -f  # Install dependencies

3. Validate installation (optional):

   ./validate-package.sh

4. Launch the application:

   • From application menu: Search for "Voice Training Data Creator"
   • From terminal: voice-training-data-creator

See INSTALL.md for detailed installation instructions and troubleshooting.

Option 2: Run from Source

For development or if you prefer not to install system-wide:

Prerequisites

• Python 3.10 or higher
• Ubuntu Desktop (tested on Ubuntu 25.04 with KDE)
• System packages: libportaudio2
• OpenAI API key (for text generation)

Setup

1. Clone the repository:

git clone https://github.com/yourusername/Voice-Training-Data-Creator.git
cd Voice-Training-Data-Creator

2. The application uses uv for virtual environment management. If you don't have it installed:

pip install uv

3. Create and activate virtual environment, then install dependencies:

uv venv .venv
source .venv/bin/activate
uv pip install -r requirements.txt

Usage

Running the Application

Use the provided launcher script:

./run.sh

Or manually:

source .venv/bin/activate
python src/main.py

First-Time Setup

1. Open Settings Tab: Click on the "Settings" tab
2. Configure Base Path: Click "Browse" to set a base directory for storing samples
3. Add API Key: Enter your OpenAI API key in the API Configuration section
4. Test Connection: Click "Test Connection" to verify your API key works
5. Set Preferred Microphone (optional): Choose your preferred microphone device
6. Set Training Goal (optional): Enter target duration in minutes (e.g., 60 for 1 hour)
7. Click "Save Settings"

Creating Samples

1. Generate Text (or use existing):

   • Set target duration (minutes)
   • Choose words per minute (WPM)
   • Select a style
   • Optionally add custom vocabulary
   • Click "Generate Text"
   • Edit if needed, or click "Regenerate" for a new version

2. Record Audio:

   • Your preferred microphone is automatically selected
   • Click "Record" to start
   • Use "Pause/Resume" if you need a break
   • Click "Retake" to discard and restart immediately
   • Click "Stop" when finished

3. Review and Save:

   • Edit the text if needed
   • Click "View Text" for a large narration view
   • Click "Save Sample" to store your recording

4. Start Next Sample:

   • Click "New Sample" to clear everything and start fresh
   • Or enable "Auto-generate next sample" in Settings for automatic workflow

Workflow Tips

• New Sample Button: Clears text and audio, optionally auto-generates new text
• Retake Button: While recording, instantly discard and restart (no stop → delete needed)
• Regenerate Button: Keep your audio but generate new text with same parameters
• Auto-generate: Enable in Settings to automatically generate text after saving
• Narration View: Use "View Text" for a large, readable display while recording

Directory Structure

Samples are organized as follows:

{base_path}/
└── samples/
    ├── 001/
    │   ├── 1.wav              # Audio recording
    │   ├── 1.txt              # Source text
    │   └── 1_metadata.json    # Generation parameters
    ├── 002/
    │   ├── 2.wav
    │   ├── 2.txt
    │   └── 2_metadata.json
    └── ...

Configuration

Configuration is stored in ~/.config/VoiceTrainingDataCreator/config.json

API keys are stored securely in the system keyring.

Troubleshooting

No Audio Devices Found

• Ensure your microphone is connected
• Check system audio settings
• Verify permissions for microphone access

API Connection Fails

• Verify your API key is correct
• Check internet connectivity
• Ensure you have sufficient OpenAI credits

Audio Clipping Warning

• Reduce microphone input volume in system settings
• Move further from the microphone
• Adjust microphone gain if available

Long Silence Detected

• Re-record the sample
• Check microphone is not muted
• Verify microphone is selected correctly

Development

Project Structure

Voice-Training-Data-Creator/
├── src/
│   ├── main.py              # Application entry point
│   ├── audio/               # Audio recording modules
│   │   ├── recorder.py
│   │   └── device_manager.py
│   ├── llm/                 # LLM integration
│   │   ├── generator.py
│   │   └── prompt_builder.py
│   ├── storage/             # Configuration and data storage
│   │   ├── config.py
│   │   └── sample_manager.py
│   ├── main.py              # Flet UI (single-file architecture)
│   └── utils/               # Validation utilities
│       └── validators.py
├── requirements.txt
├── run.sh                   # Launcher script
└── README.md

Dependencies

• Flet: Modern GUI framework (Flutter-based)
• sounddevice: Audio recording
• soundfile: WAV file handling
• numpy: Audio processing
• openai: LLM text generation
• keyring: Secure credential storage

UI Framework

This application uses Flet (Flutter-based) for the GUI, providing:

• Flutter-based rendering with predictable layouts
• Declarative UI architecture
• Better cross-platform support
• Material Design by default
• Three-tab interface for organized workflow

The app was migrated from PyQt6 to Flet to resolve persistent layout issues and provide a more modern user experience.

License

MIT