PHONOSCRIPT

Inspiration

• Screenwriters struggle to hear scripts before production; table reads are expensive and scheduling cast is painful.
• We wanted a 24/7 “audio rehearsal room” where writers can upload a PDF and instantly hear an expressive performance.

What it does

• Parses uploaded screenplays with the AI/ML API, extracting scenes, characters, and dialogue.
• Lets users cast ElevenLabs voices, generate scene audio, and trigger instant table reads in the dashboard.
• Supports multilingual dubbing, audio export, and Supabase-backed persistence for collaborative teams.

How we built it

• React + Vite frontend with Tailwind glassmorphism UI and Supabase auth/state.
• Hono edge backend on Vercel that calls the AI/ML API for script parsing and ElevenLabs for TTS via fetch-compatible handlers.
• Supabase Postgres + storage for script assets, characters, and metadata.

Challenges we ran into

• Making ElevenLabs calls Edge-compatible (axios was unsupported; we rewrote with native fetch + Uint8Array handling).
• Keeping large text parsing fast while staying within AI/ML API context limits.

Accomplishments that we're proud of

• Fully automated, voice-casted table reads that sound production ready.
• Clean separation between parsing, storage, and playback that scales for multiple users.
• Polished landing/dashboard experience that demos the product without manual setup.

What we learned

• Edge runtimes demand fetch-native integrations; legacy Node clients often fail.
• Prompt engineering for screenplay structure requires tight JSON schemas and defensive parsing.
• Supabase storage + database combo is a strong fit for media-heavy Next/Vite apps.

What's next

• Add real-time co-writing with synchronized playback and commenting.
• Introduce emotion-aware fine-tuning for ElevenLabs voices per character arcs.
• Roll out pricing tiers with usage analytics and self-serve billing.

Built With

• css
• elevenlabs
• node.js
• react
• supabase
• tailwind
• vite