GenVoice Studio

AI-powered voice cloning & generation from text — built on Kaggle GPU infrastructure.

GenVoice Studio lets you clone a voice or design a new one and synthesize any text at broadcast quality, all from a seamless, mobile-responsive web interface. No GPU required on your end — generation runs on a free Kaggle T4 GPU via their API.

Powered by Breeze TTS 2.

🔥 Try GenVoice Studio Live Now! 🔥

---

💸 The Ultimate Free Alternative to ElevenLabs

GenVoice Studio is an entirely FREE, practically UNLIMITED voice cloning, text-to-speech, and AI story writing studio.

Unlike ElevenLabs and other corporate TTS services that trap you behind:
❌ Pay-per-character pricing tiers
❌ Strict token quotas and rate limits
❌ Expensive proprietary subscriptions

GenVoice Studio bypasses the paywalls completely by tapping into Kaggle's Free GPU and AI Credits. By simply linking a free Kaggle account, you unlock:
✅ 100% Free broadcast-quality voice cloning and TTS generation
✅ Practically Unlimited characters (up to 30 free T4 GPU hours per week!)
✅ Free AI Story Generation & Script Optimization using 20+ state-of-the-art LLMs (DeepSeek R1, Gemini 3, GPTs, Claude)
✅ Zero-Cost Operation — host the front-end for free on Vercel and let Kaggle handle the heavy GPU lifting!

---

🌟 Key Features

• Voice Clone & Voice Design
  • Voice Clone: Upload any custom audio clip (WAV/MP3) to instantly clone a speaker's voice.
  • Voice Design: Create an entirely new voice from scratch simply by describing it (e.g., "A raspy old man with a British accent").
• ✨ AI Write Story
  • Need a script? Use the built-in AI story generator. Pick a topic, tone (e.g., Dramatic, Funny), and length, and it uses Kaggle's LLM endpoints to write an expressive script with vocal cues like (laugh) and (sigh).
• Smart Generation Routing & Chunking
  • Short text is generated rapidly in a single pass.
  • Long text is automatically chunked at the sentence level to prevent GPU out-of-memory errors.
• Studio-Grade Editor
  • Live progress bars and ETA countdowns for ongoing generations.
  • Granular chunk editor: Switch between Crossfade (smooth blending) and Trim (hard cut) for individual sentences, and preview immediately.
  • Result Mode: Once audio is generated, the UI cleanly transforms into a focused result player, with quick options to Retry or start a New Script.
• BYOK (Bring Your Own Kaggle) & Admin Auth
  • Users can authenticate directly on the site using their own Kaggle Username/Key (no server config required).
  • Site owners can use an Admin Password to fall back to server-configured .env credentials.
• Serverless & Resilient
  • Fully serverless frontend and middleware hosted on Vercel.
  • Heavily optimized polling mechanism with transient error resilience (transparently handles Kaggle 503s during long-running notebook sessions).

---

🏗️ Stack & Architecture

Layer	Technology
Frontend	Next.js 16 (App Router), React, Tailwind CSS
Backend	Next.js API Routes (Edge-compatible)
AI Model	Breeze TTS 2
Compute Engine	Kaggle Notebooks API (T4 x2) + Kaggle LLM API
Deployment	Vercel

How It Works

1. Authentication: The frontend mints an OAuth model proxy token via Kaggle's API using the user's (or admin's) credentials.
2. Execution: The /api/generate route pushes a dynamic Jupyter Notebook to Kaggle via the kernel-push API.
3. Compute: A Kaggle T4 GPU spins up, installs dependencies, loads Breeze TTS 2, processes the instructions (cloning or designing), and generates the WAV outputs.
4. Resilient Polling: The /api/status route persistently polls Kaggle's status endpoints. It intelligently ignores transient network drops (like 503s/504s) and parses live progress.
5. Retrieval: Upon completion, the API fetches the final .wav outputs (or chunks) directly from Kaggle's artifact storage and delivers them to the user.

---

🚀 Local Development

Prerequisites

• Node.js 18+
• A Kaggle account with API access enabled
• A Kaggle API key (kaggle.json → username + key)

Setup

git clone https://github.com/pundhiranshul/genVoice.git
cd genVoice
npm install

Create a .env.local file:

# Fallback credentials used when logging in via 'Admin Mode'
KAGGLE_USERNAME=your_kaggle_username
KAGGLE_TOKEN=your_kaggle_api_key
KAGGLE_KERNEL_SLUG=genvoice-api
APP_PASSWORD=your_chosen_admin_password

Run the development server:

npm run dev

Open http://localhost:3000.

---

☁️ Deployment (Vercel)

1. Push your repository to GitHub.
2. Import the project in Vercel.
3. Add the environment variables (KAGGLE_USERNAME, KAGGLE_TOKEN, KAGGLE_KERNEL_SLUG, APP_PASSWORD) in your Vercel project settings.
4. Deploy!

---

⚠️ Limitations

• Generation time: Kaggle container startup + model loading takes roughly ~2 minutes. Actual inference takes an additional 1-5+ minutes depending on text length.
• Kaggle Quotas: Free Kaggle tiers provide 30 GPU hours/week per account.
• API Timeouts: Next.js Serverless functions time out after 10–15s on Vercel's free tier. GenVoice avoids this by strictly separating the trigger (/api/generate) from the asynchronous status loop (/api/status).

---

📄 License

MIT License. See LICENSE for details.