Product 03
Voice,
on your terms.
speech is a compact voice service: text to speech and speech to text behind an OpenAI-compatible API. The browser talks to your server — and only to your server. Keys never reach the client.
Drop-in endpoint
POST /v1/audio/speech — a drop-in for anything that already talks the OpenAI audio API. Point the base URL at your instance and you are done.
Streams while it thinks
24 kHz audio starts playing during generation — no waiting for the finished file. A regular WAV is written alongside for replay and download.
Custom voices
Custom voices with language and style instructions — warm and calm for the meeting room, brisk for notifications, yours to define.
Transcription included
Whisper speech-to-text with incremental results streamed as they form — upload a file, drag it in, or record straight in the browser.
A thin, honest proxy
The service itself needs no GPU: models run on your vLLM or GPUStack endpoint, and speech stays a small, auditable app in front of them.
Keys stay home
The browser only ever talks to the app. Provider keys live in server-side configuration and are never shipped in HTML or JavaScript.