STT

Overview

Speech-to-text transcription through the environment’s STT provider, with usage tracking.

Supported providers include OpenAI, Deepgram, AssemblyAI, Google, Azure, and Athena (OpenAI-compatible Live API via tunnel or custom baseUrl ending in /v1).

API

{pe} = /projects/{projectId}/environments/{environmentId}.

Method	Path	Notes
GET	`/stt/health`	Health
POST	`/stt{pe}/transcribe`	RequireAuthorization — multipart `file`, returns JSON
POST	`/stt{pe}/transcribe/stream`	RequireAuthorization — multipart `file`, SSE transcript deltas
GET	`/stt{pe}/usage`	RequireAuthorization

Streaming events

POST /transcribe/stream returns text/event-stream:

data: {"type":"transcript.text.delta","delta":"…","text":"…"}
data: {"type":"transcript.text.done","text":"…"}
data: [DONE]

Athena uses native upstream SSE (stream=true). Other providers run a batch transcription and emit a synthetic delta + done.

Clients

Package	Runtime
`@vijihub/stt`	server
`VijiHub.STT`	.NET

import { createSttClient } from '@vijihub/stt'

const stt = createSttClient({
  baseUrl: 'http://localhost:5600',
  projectId: '...',
  environmentId: '...',
  getAccessToken: () => auth.ensureAccessToken(),
})

Implementation

Enable STT and configure the provider (and optional tunnel URL).
For Athena, set baseUrl to https://{slug}.tunnel.vijihub.com/v1 (API key optional).
POST audio to /transcribe or /transcribe/stream from a trusted client with Bearer auth.
Track consumption via /usage.

Constraints

Requires authorization — keep calls on server/BFF.
Audio size/format limits depend on the configured provider (Athena prefers WAV).
Provider credentials stay in environment settings.