AzureTTS

Use AzureTTS when you want Microsoft Azure’s catalog of hundreds of neural voices across 140+ locales via a simple REST call. Each synthesize() request POSTs an SSML document to your regional Speech endpoint and returns the complete audio as a Blob — no WebSocket management required. Speaking styles (cheerful, newscast, …) and rate/pitch controls are exposed as config options.

Prerequisites

• An Azure Speech resource (key + region) or a CompositeVoice proxy server
• No additional dependencies required. AzureTTS uses native fetch internally.

Basic setup

import { CompositeVoice, NativeSTT, AnthropicLLM, AzureTTS, BrowserAudioOutput } from 'composite-voice';

const voice = new CompositeVoice({
  providers: [
    new NativeSTT(),
    new AnthropicLLM({
      proxyUrl: '/api/proxy/anthropic',
      model: 'claude-haiku-4-5',
    }),
    new AzureTTS({
      proxyUrl: '/api/proxy/azure-tts',
      voiceName: 'en-US-AriaNeural',
      outputFormat: 'audio-24khz-48kbitrate-mono-mp3',
    }),
    new BrowserAudioOutput(),
  ],
});

await voice.initialize();
await voice.startListening();

Configuration options

Available voices

List voices with GET https://<region>.tts.speech.microsoft.com/cognitiveservices/voices/list (send your key as Ocp-Apim-Subscription-Key), or browse the voice gallery. Voice names look like en-US-AriaNeural, en-GB-SoniaNeural, or de-DE-KatjaNeural. Some voices support speaking styles — check the voice’s StyleList.

Output formats

Complete example

import { CompositeVoice, MicrophoneInput, DeepgramSTT, AnthropicLLM, AzureTTS, BrowserAudioOutput } from 'composite-voice';

const tts = new AzureTTS({
  proxyUrl: '/api/proxy/azure-tts',
  voiceName: 'en-US-AriaNeural',
  style: 'cheerful',
  rate: 1.1,
});

const voice = new CompositeVoice({
  providers: [
    new MicrophoneInput(),
    new DeepgramSTT({ proxyUrl: '/api/proxy/deepgram' }),
    new AnthropicLLM({
      proxyUrl: '/api/proxy/anthropic',
      model: 'claude-haiku-4-5',
    }),
    tts,
    new BrowserAudioOutput(),
  ],
});

voice.on('tts.start', () => console.log('Speaking...'));
voice.on('tts.end', () => console.log('Done speaking'));

await voice.initialize();
await voice.startListening();

Authentication modes

• Subscription key (direct): pass your Speech resource key as a string apiKey — it is sent as the Ocp-Apim-Subscription-Key header.
• Bearer token (direct): pass an async apiKey factory that fetches a token from your server (which exchanges the key at POST https://<region>.api.cognitive.microsoft.com/sts/v1.0/issueToken). Tokens are valid for 10 minutes and are sent as Authorization: Bearer; the factory is called on every request so refreshed tokens are picked up automatically.
• Proxy (recommended): pass proxyUrl and configure azureSpeechApiKey + azureSpeechRegion in your proxy — the key never reaches the browser.

Tips

• User text is XML-escaped automatically before being embedded in the SSML <speak> document, so &, <, and quotes in LLM output are safe.
• AzureTTS is REST-based, not streaming. The full audio Blob is returned after the API processes the entire input. For real-time streaming, consider DeepgramTTS.
• Synthesis is capped at 10 minutes of audio per request by the service.
• If you select a 48 kHz output format, the high-fidelity 48 kHz voice model is invoked automatically.

Further reading

• API reference
• Providers reference
• Getting started