Voicey — AI-Powered Text-to-Speech & Voice Cloning Platform
A full-stack SaaS application that lets users generate lifelike speech from text, clone voices from audio samples, and manage voice libraries with AI-powered TTS on serverless GPUs.
Role
Full-Stack Engineer
Outcome
Full-stack AI SaaS with organization-based multi-tenancy, pay-as-you-go billing with metered usage, serverless GPU inference for voice cloning, and real-time audio recording and playback
Team
Solo Developer
Overview
Building the product from concept to execution
Voicey is a comprehensive AI-powered Text-to-Speech and Voice Cloning SaaS application built with Next.js 16 (App Router) and React 19. The platform enables users to generate lifelike speech from text, clone voices from audio samples, and manage voice libraries across 12 categories. Key features include a 20-system-voice library with procedural DiceBear avatars, in-browser audio recording via RecordRTC, adjustable TTS parameters (temperature, top-P, top-K, repetition penalty), real-time cost estimation at $0.0003 per character, and WaveSurfer.js waveform visualization. The application implements organization-based multi-tenancy with Clerk auth, pay-as-you-go billing via Polar.sh with metered usage tracking, Cloudflare R2 for audio storage with 1-hour signed URLs, and Sentry for error monitoring across edge and server runtimes. The TTS engine leverages Chatterbox TTS running on Modal's serverless GPU platform (A10G) for scalable inference. Full type safety is maintained end-to-end via tRPC v11, Prisma ORM v7, and TypeScript, with TanStack Form for type-safe form management and TanStack Query for server state caching.
Highlights
Key achievements
- Full-stack AI SaaS with organization-based multi-tenancy, pay-as-you-go billing with metered usage, serverless GPU inference for voice cloning, and real-time audio recording and playback
- Designed and implemented a full-stack experience with strong attention to usability, structure, and maintainability.
- Built with a focus on responsive behavior, performance, and production-ready execution.
