category:
  id: voice
  label: Voice
  description: Voice & speech synthesis
  defaultPlatform: elevenlabs
  defaultMode: detailed

# Currency policy (1.16.0): entries touched by a fact-verification pass carry
# `verifiedAt`. Entries without a date have not been re-verified recently.

platforms:
  - id: elevenlabs
    label: ElevenLabs (v3)
    description: Eleven v3 — expressive TTS with audio tags and emotion control; voice cloning from 10s of audio; Flash/Turbo v2.5 for low latency
    verifiedAt: "2026-10-05"
    source: elevenlabs.io
    syntaxHints:
      - 'audio tags in v3: [laughs], [whispers], [excited], [sad] — inline emotion/sound direction'
      - voice settings: stability, similarity, style (v2 API); v3 uses tag-driven expressiveness
      - conversational v3 model for real-time agents
      - voice cloning — instant (seconds) or professional
      - 32+ languages (v2 multilingual)
      - pronunciation control via respellings/phonetics

  - id: openai-tts
    label: OpenAI TTS (gpt-4o-mini-tts)
    description: Instruction-following voices — steer tone and delivery with natural-language directions; legacy alloy/echo/nova set still available
    verifiedAt: "2026-10-05"
    syntaxHints:
      - 'natural-language voice instructions (steer tone, pacing, emotion: "speak like a sympathetic customer-service agent")'
      - 'voices: the gpt-4o-mini-tts instruction-following set (plus legacy alloy/echo/fable/onyx/nova/shimmer)'
      - speed adjustment (0.25-4.0)
      - no SSML — describe the delivery in words

  - id: fish-audio
    label: Fish Audio
    description: Voice cloning, multilingual
    syntaxHints:
      - voice cloning
      - multilingual
      - emotional control
      - reference audio

  - id: sesame
    label: Sesame
    description: Conversational AI voices
    syntaxHints:
      - conversational style
      - emotional expression
      - natural dialogue
      - character voices

  - id: google-tts
    label: Google TTS
    description: Cloud TTS, WaveNet voices
    syntaxHints:
      - SSML support
      - WaveNet voices
      - Neural2 voices
      - speaking rate
      - pitch

  - id: playht
    label: PlayHT
    description: Ultra-realistic TTS, voice cloning
    syntaxHints:
      - voice selection
      - emotion control
      - speed
      - voice cloning
      - PlayHT 2.0 turbo

  - id: kokoro
    label: Kokoro
    description: Open weights, fast expressive TTS
    syntaxHints:
      - open weights
      - voice presets
      - emotional control
      - fast inference
      - local deployment