Build on the supported package contract
Use @absolutejs/voice-azure through its supported public entry points.
@absolutejs/voice-azure
@absolutejs/voice-azurev0.0.1-beta.7betaVoice & MediaAzure Speech (Cognitive Services) adapter for @absolutejs/voice — Neural TTS over REST + streaming STT over the WebSocket Unified Speech Protocol
#Installation
BASH
bun add @absolutejs/voice-azure#Capabilities
Overview
Azure Speech (Cognitive Services) adapter for @absolutejs/voice — Neural TTS over REST plus streaming STT over Azure's WebSocket Unified Speech Protocol.
TTS
For telephony bridges, use a μ-law raw format at 8 kHz:
Bearer-token auth (10-minute Azure auth tokens) is also supported:
Options
Option — Required — Default — Notes
subscriptionKey / token — one of — — — Subscription key sent as Ocp-Apim-Subscription-Key, or short-lived bearer token sent as Authorization.
voice — yes — — — Azure voice name, e.g. en-US-JennyNeural, fr-FR-DeniseNeural.
Show 10 more
region — yes\ — — — Azure region (eastus, westus, francecentral, …). \ Or pass baseUrl directly.
baseUrl — no — https://{region}.tts.speech.microsoft.com — Override for sovereign clouds or Azure private endpoints.
endpointPath — no — /cognitiveservices/v1 — Override if you front the service with a gateway.
outputFormat — no — raw-24khz-16bit-mono-pcm — Must be a raw- format (mp3/wav variants are rejected because they aren't streamable frame-by-frame).
language — no — en-US — Used in the SSML xml:lang attribute.
voiceStyle — no — — — Azure neural style (cheerful, empathetic, customerservice, …).
styleDegree — no — — — Only applied when voiceStyle is set (0..2 typically).
prosody — no — — — { rate, pitch, volume } — strings forwarded to the SSML element.
userAgent — no — @absolutejs/voice-azure — Sent as User-Agent.
fetch — no — globalThis.fetch — Inject for tests; opportunistic HTTP/2 multiplexing is enabled for HTTPS targets.
Notes
Only raw- output formats are supported because the voice runtime needs framed PCM/μ-law/α-law to feed transports without buffering the whole response. If you need MP3/WAV for offline assets, call the Azure REST API directly.
The adapter aborts the in-flight HTTP request on session.close(reason) and refuses further send() calls.
Whitespace-only send() is a no-op (matches the ElevenLabs and Cartesia adapters).
Show 1 more
Bearer tokens expire after 10 minutes by default — refresh externally and pass the new value into a fresh adapter, or stick with subscriptionKey for long-running deployments.
STT
The STT adapter speaks Azure's WebSocket Unified Speech Protocol directly (no Microsoft SDK dependency):
Connects to wss://{region}.stt.speech.microsoft.com/speech/recognition/{mode}/cognitiveservices/v1
Authenticates via Ocp-Apim-Subscription-Key or Authorization: Bearer header.
Show 10 more
Sends a speech.config text frame with system metadata on open, then queues any audio sent before the socket finished opening.
Prepends a 44-byte RIFF/WAV header to the first audio chunk (using the format declared in STTAdapterOpenOptions.format) and ships subsequent chunks as raw PCM under the same audio/x-wav content type.
Maps speech.hypothesis → partial events, speech.phrase → final events (RecognitionStatus === "Success" only, with NBest[0].Confidence lifted onto the transcript), and turn.end → endOfTurn with reason: "vendor".
STT options
Option — Required — Default — Notes
subscriptionKey / token — one of — — — Same auth choices as TTS.
region — yes\ — — — Azure region. \ Or pass baseUrl.
baseUrl — no — wss://{region}.stt.speech.microsoft.com — Override for sovereign clouds, private endpoints, or test stubs.
endpointPath — no — /speech/recognition/{mode}/cognitiveservices/v1 — Override if you front the service with a gateway.
recognitionMode — no — conversation — conversation / dictation / interactive.
Outcomes
What you can build
Hardening checklist
Production guidance
Make every external boundary explicitPin the deployed @absolutejs/voice-azure version, replace example or memory-backed dependencies with durable implementations, bound external calls, protect credentials, and emit enough evidence to retry or recover safely.
Follow in order
Troubleshooting path
1
Trace from the first failed boundary
Reproduce the smallest canonical @absolutejs/voice-azure example, confirm the supported entry point and version in the API explorer, then inspect the first boundary that did not produce its documented result.
#TTS
Partial snippetWorking example for TTS.
TS
import { voice } from "@absolutejs/voice";
import { azureTTS } from "@absolutejs/voice-azure";
const app = voice({
// ... stt + other voice options ...
tts: azureTTS({
region: "eastus",
subscriptionKey: process.env.AZURE_SPEECH_KEY!,
voice: "en-US-JennyNeural",
// optional:
outputFormat: "raw-24khz-16bit-mono-pcm", // default
language: "en-US", // default
voiceStyle: "cheerful",
styleDegree: 1.5,
prosody: { rate: "fast", pitch: "+5%" },
}),
});#TTS 2
Partial snippetFor telephony bridges, use a μ-law raw format at 8 kHz:
TS
azureTTS({
region,
subscriptionKey,
voice: "en-US-AriaNeural",
outputFormat: "raw-8khz-8bit-mono-mulaw",
});#Public entry points
Supported entry points declared by this package manifest.
Package entry point declared in package.json.
@absolutejs/voice-azure@absolutejs/voice-azure/manifest@absolutejs/voice-azure/manifest.json
#Package commands
Scripts declared by this package manifest.
bun run buildrm -rf dist && bun build ./src/index.ts ./src/manifest.ts --outdir dist --target bun --external @absolutejs/manifest --external '@absolutejs/manifest/*' --external @absolutejs/voice --external '@absolutejs/voice/*' --external @sinclair/typebox --external '@sinclair/typebox/*' && tsc --emitDeclarationOnly --project tsconfig.json && absolute-manifest emit
bun run check:packagebun run format && bun run typecheck && bun run test && bun run verify-package && bun run build && bun run verify-package --artifacts
bun run formatprettier --write "./**/*.{js,ts,json,md}"
bun run testbun test
bun run typecheckbun run tsc --noEmit