ContextQMD
Back to Promptfoo

provider-elevenlabs/tts (ElevenLabs Text-to-Speech)

examples/provider-elevenlabs/tts/README.md

0.121.92.1 KB
Original Source

provider-elevenlabs/tts (ElevenLabs Text-to-Speech)

You can run this example with:

bash
npx promptfoo@latest init --example provider-elevenlabs/tts
cd provider-elevenlabs/tts

Test and compare ElevenLabs TTS models and voice settings.

What this tests

  • Model comparison: Flash v2.5, Turbo v2.5, Multilingual v2
  • Streaming vs. non-streaming performance
  • Voice quality across different text inputs
  • Cost and latency metrics

Setup

Set your ElevenLabs API key:

bash
export ELEVENLABS_API_KEY=your_api_key_here

Run the example

bash
npx promptfoo@latest eval -c ./promptfooconfig.yaml

Or view in the UI:

bash
npx promptfoo@latest eval -c ./promptfooconfig.yaml
npx promptfoo@latest view

What to look for

  1. Model differences: Flash v2.5 has lowest latency (~200ms), Multilingual v2 best quality
  2. Streaming benefits: First chunk arrives in ~75ms for real-time feel
  3. Cost tracking: ~$0.02 per 1000 characters
  4. Audio metadata: Duration, size, format info in response

Available voices

This example uses Rachel (21m00Tcm4TlvDq8ikWAM). Try other popular voices:

  • Rachel: Calm, clear female voice (default)
  • Clyde: Warm, grounded male voice (2EiwWnXFnvU5JabPnv8n)
  • Drew: Well-rounded male voice (29vD33N1CtxCmqQRPOHJ)
  • Paul: Casual male voice (5Q0t7uMcjvnagumLfvZi)

Voice settings

Customize the voice output:

yaml
voiceSettings:
  stability: 0.5 # 0 (more variable) to 1 (more stable)
  similarity_boost: 0.75 # 0 (low) to 1 (high)
  style: 0.0 # 0 to 1 (only for v2 models)
  use_speaker_boost: true # Enhance clarity
  speed: 1.0 # 0.25 to 4.0

Output formats

Available formats:

  • mp3_22050_32 - Smallest size, lower quality
  • mp3_44100_128 - Balanced (default)
  • mp3_44100_192 - High quality
  • pcm_16000 - Raw PCM for processing
  • pcm_44100 - High quality PCM
  • ulaw_8000 - Phone quality

Learn more