Dream TTS turned local speech from a model demo into product architecture: Flutter for the surface, Rust for durable synthesis work, and strict boundaries for long-running private …
Local speech becomes trustworthy when pronunciation, latency, fixtures, logs, and release gates make quality repeatable instead of subjective.
A local speech feature survives only when the app talks to one stable product API while engines, voices, platforms, playback, and failures change underneath it.
Your first local voice ships only after it is bundled, checksummed, observable, cancellable, and boring enough to survive real user machines.
If your app needs speech without a hosted API, the hard part comes after the nice demo voice: engines, latency budgets, licenses, updates, and fallbacks.