Text-to-Speech API

The ElevenLabs Text-to-Speech API is the core tool for developers. From simple REST calls to streaming audio with sub-300ms latency — here you'll learn everything you need for integration.

REST API Basics

Simple TTS Call

const response = await fetch(
  'https://api.elevenlabs.io/v1/text-to-speech/21m00Tcm4TlvDq8ikWAM',
  {
    method: 'POST',
    headers: {
      'xi-api-key': process.env.ELEVENLABS_API_KEY,
      'Content-Type': 'application/json',
    },
    body: JSON.stringify({
      text: 'Welcome to our service.',
      model_id: 'eleven_multilingual_v2',
      voice_settings: {
        stability: 0.5,
        similarity_boost: 0.75,
        style: 0.3,
        use_speaker_boost: true,
      },
    }),
  }
)
const audioBuffer = await response.arrayBuffer()

Voice Settings Explained

Model Selection

Streaming Audio

Why Streaming?

With the normal API, the client waits until the entire audio is generated. With streaming, the first bytes arrive in < 300 ms — critical for real-time applications.

const response = await fetch(
  'https://api.elevenlabs.io/v1/text-to-speech/21m00Tcm4TlvDq8ikWAM/stream',
  {
    method: 'POST',
    headers: {
      'xi-api-key': process.env.ELEVENLABS_API_KEY,
      'Content-Type': 'application/json',
    },
    body: JSON.stringify({
      text: 'This will be returned as a stream.',
      model_id: 'eleven_turbo_v2_5',
      output_format: 'mp3_44100_128',
    }),
  }
)

// Forward stream directly to client
const reader = response.body.getReader()
while (true) {
  const { done, value } = await reader.read()
  if (done) break
  // Process audio chunks
}

Output Formats

SSML Support

Speech Synthesis Markup Language

SSML gives you fine-grained control over speech output:

<speak>
  Welcome to <emphasis level="strong">EverStrategy.ai</emphasis>.
  <break time="500ms"/>
  Your appointment is on <say-as interpret-as="date">2026-03-15</say-as>.
  That costs <say-as interpret-as="currency">49.99 EUR</say-as>.
</speak>

Supported SSML Tags

• <break> — Insert pause (time="500ms")
• <emphasis> — Emphasis (level: strong, moderate, reduced)
• <say-as> — Interpretation: date, currency, telephone, cardinal
• <phoneme> — Exact pronunciation via IPA
• <prosody> — Speed, pitch, volume

Latency Optimization

The Latency Formula

Total latency = API overhead + Model inference + Text length + Network

Optimization Strategies

1. Use Turbo model — eleven_turbo_v2_5 instead of multilingual_v2
2. Enable streaming — first bytes in < 300 ms
3. Chunk text — split long texts into sentences and send in parallel
4. Use CDN — cache generated audio and serve via CDN
5. Regional endpoints — server location close to the user
6. Pre-generation — pre-generate and cache common announcements

Latency Benchmarks

Practical tip: For real-time voice agents, always use the Turbo model with streaming. For offline production (audiobooks, podcasts), use the Multilingual v2 model — the higher quality is worth the longer wait.