Voice Coach

A simple real-time voice analysis tool using Vosk, Parselmouth, Librosa, and Electron to provide feedback on speaking habits like pitch (F0), volume (RMS), upward inflection, staccato rhythm, and vocal shakiness (jitter/shimmer).

This is a pet project primarily developed for macOS.

Features

• Real-time calculation of RMS (volume) and F0 (pitch).
• Utterance-level analysis of:
  • Jitter & Shimmer (using Praat via Parselmouth)
  • Staccato Rhythm (based on pause/word duration statistics from Vosk)
  • Upward Inflection (based on F0 slope at utterance end)
• Simple Electron UI displaying metrics and visual alerts.

Tech Stack

• Backend: Python 3
• Speech-to-Text: Vosk
• Acoustic Analysis: Parselmouth (Praat), Librosa
• Audio I/O: sounddevice
• Frontend: Electron, Node.js

Setup Instructions

1. Clone Repository:

   git clone <your-repo-url>
   cd voicecoach

2. Python Setup (Requires Python 3.9+):

   • Create a virtual environment:

     python3 -m venv venv

   • Activate the environment:
     • macOS/Linux: source venv/bin/activate
     • Windows: .\venv\Scripts\activate
   • Install Python dependencies:

     pip install -r requirements.txt

3. Download Vosk Model:

   • Download the model vosk-model-small-en-us-0.15 from https://alphacephei.com/vosk/models.
   • Extract the downloaded archive.
   • IMPORTANT: Place the extracted folder (which should be named vosk-model-small-en-us-0.15) directly into the root directory of this project (voicecoach/).

4. Node.js Setup (Requires Node >= v22, npm >= v10):

   • Install Node.js and npm if you haven't already: https://nodejs.org/
   • Install Node dependencies:

     npm install

Running the App

Ensure your Python virtual environment is deactivated. From the project's root directory (voicecoach/), run:

npm start

The application window should open and automatically start listening.

Key Files

• voice.py: Python backend script handling audio capture, analysis, and JSON output.
• main.js: Electron main process script, manages the app window and Python child process.
• preload.js: Electron preload script for secure IPC.
• renderer.js: Electron renderer process script, handles UI logic and updates.
• index.html: Defines the UI structure.

Known Issues/Limitations

• Primarily tested on macOS.
• Shakiness detection uses basic Jitter/Shimmer thresholds that may need tuning (renderer.js).
• Staccato detection rules are experimental (voice.py).
• Requires the specific Vosk model vosk-model-small-en-us-0.15 placed in the root folder.