Web Applications / Project 10
Vocable Studio
An offline speech studio with local synthesis engines and a shared typed core.
01 / Overview
Behind the interface.
An offline text to speech studio that runs entirely on the user's own machine with no cloud, account or API key. Four synthesis engines sit behind one typed provider contract: two neural models running in WebAssembly through ONNX Runtime, the operating system's own voices, and a formant synthesiser that ships inside the app. The only moment it touches the network is installing a voice, and a privacy setting can block even that.
02 / Implementation
What makes it work.
A dependency free Rust domain crate is the single source of truth for the type system, projected into 83 TypeScript types through ts-rs with a check mode that fails when the committed bindings drift.
One core interface of sixty four methods with a live in process implementation, plus a second complete and compiler checked IPC client waiting on its Rust backend.
Both neural engines share a full multi language espeak-ng front end compiled to WebAssembly, batched so an entire script is phonemised in one instance rather than remounting eighteen megabytes of pronunciation data per call.
Model installs stream real byte level progress, resume through range requests, verify against the publisher's MD5 before committing, record a SHA-256 for later checks, and land in one atomic IndexedDB transaction.
A voice library of 178 Piper voices across 53 languages plus 28 Kokoro voices, with sideloading for air gapped installs.
454 TypeScript tests across twenty seven files and 169 Rust tests, including a conformance suite every synthesis provider has to pass.