SmallestTTS

Use SmallestTTS when you want ultra-low-latency speech synthesis from Smallest.ai’s Lightning models via a simple REST call. Each synthesize() request hits the unified Waves POST /waves/v1/tts endpoint and returns the complete audio as a Blob — no WebSocket management required. Voices from the Waves catalog and cloned voices are both supported through the voiceId option.

Prerequisites

• A Smallest.ai API key or a CompositeVoice proxy server
• No additional dependencies required. SmallestTTS uses native fetch internally.

Basic setup

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

const voice = new CompositeVoice({
  providers: [
    new NativeSTT(),
    new AnthropicLLM({
      proxyUrl: '/api/proxy/anthropic',
      model: 'claude-haiku-4-5',
    }),
    new SmallestTTS({
      proxyUrl: '/api/proxy/smallest',
      voiceId: 'meher',
      model: 'lightning_v3.1',
      outputFormat: 'wav',
    }),
    new BrowserAudioOutput(),
  ],
});

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

Configuration options

Available voices

Browse and preview voices in the Waves console, or clone your own. The quickstart voices meher, magnus, and olivia are good starting points.

Output formats

Complete example

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

const tts = new SmallestTTS({
  proxyUrl: '/api/proxy/smallest',
  voiceId: 'meher',
  model: 'lightning_v3.1_pro',
  outputFormat: 'wav',
  sampleRate: 24000,
  speed: 1.0,
});

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();

Model selection

• lightning_v3.1 — The standard 44 kHz Lightning model with natural, expressive speech. Supports 12 languages (English, Hindi, Spanish, and 9 Indian languages) plus voice cloning. Best default.
• lightning_v3.1_pro — Curated 44 kHz voice pool with improved naturalness across American, British, and Indian accents, at the same latency.

Older Lightning models (lightning, lightning-large, lightning-v2) and their per-model get_speech endpoints are deprecated by Smallest.ai; SmallestTTS only targets the unified /waves/v1/tts endpoint.

Tips

• Smallest.ai recommends keeping each request under roughly 250 characters of text for the lowest latency.
• SmallestTTS is REST-based, not streaming. The full audio Blob is returned after the API processes the entire input. For real-time streaming, consider DeepgramTTS.
• The API’s own default output format is raw pcm; SmallestTTS defaults to wav instead so the returned Blob is directly playable in browsers.
• Use proxyUrl in production so your API key stays server-side — set smallestApiKey in your proxy config.

Further reading

• API reference
• Providers reference
• Getting started