BlitzVoice

Record pronunciation datasets and train custom TTS voice models for BlitzGrok courses

In this guide

Overview

BlitzVoice is a desktop companion app you download from the Tools Hub. It helps you build pronunciation corpora for Sanskrit and other Indic languages using IAST input, record WAV samples with your microphone, and train Piper voice models for use in BlitzGrok pronunciation-practice questions.

All recordings and project data are stored locally under ~/BlitzVoice/ until you choose to publish.

Install & Run

For macOS (Apple Silicon):

Log in as an educator or admin on BlitzGrok
Open the Tools tab on your home page
Download BlitzVoice v0.1.0 for Mac
Open the DMG and drag BlitzVoice to Applications

Requirements:

macOS 12 or later on Apple Silicon

Workflow

1. Create a project

Open BlitzVoice, create a new project, and optionally link your BlitzGrok API URL for later publish.

2. Import corpus

Paste IAST text or import CSV. Tag each item as letter, word, or sentence granularity. Devanagari preview appears automatically.

3. Capture recordings

Use the Capture screen to record prompts in random order (letters weighted higher). Save & Next advances through unrecorded items.

4. Validate

Review recordings with playback, check coverage stats by granularity, and re-record as needed.

Publish to BlitzGrok

After training a Piper model, use the Publish screen to upload your voice model to BlitzGrok using your API URL and vendor API key. Published models will power TTS in pronunciation-practice questions.

Coming soon: BlitzGrok Voice Model API integration is not yet available in this release.