Fish Audio Speech Plugin for OpenClaw

A speech provider plugin that integrates Fish Audio TTS with OpenClaw.

Features

• Fish Audio S2-Pro / S1 / S2 model support
• Dynamic voice listing — your own cloned voices + official Fish Audio voices
• Format-aware output — opus for voice notes (Telegram, WhatsApp), mp3 otherwise
• Inline directives — switch voice, speed, model, and latency mid-message
• No core changes required — standard SpeechProviderPlugin extension

Installation

openclaw plugins install @openclaw/fish-audio-speech

Configuration

In your openclaw.json:

{
  "messages": {
    "tts": {
      "provider": "fish-audio",
      "providers": {
        "fish-audio": {
          "apiKey": "your-fish-audio-api-key",
          "voiceId": "8a2d42279389471993460b85340235c5",
          "model": "s2-pro",
          "latency": "normal",
          "speed": 1.0
        }
      }
    }
  }
}

Config Options

Field	Type	Default	Description
apiKey	string	—	Required. Fish Audio API key
voiceId	string	8a2d42...	Reference ID of the voice to use
model	string	s2-pro	TTS model (s2-pro, s1, s2)
latency	string	normal	Latency mode (normal, balanced, low)
speed	number	—	Prosody speed (0.5–2.0)
temperature	number	—	Sampling temperature (0–1)
topP	number	—	Top-p sampling (0–1)
baseUrl	string	https://api.fish.audio	API base URL

Environment Variable

You can also set the API key via environment variable:

FISH_AUDIO_API_KEY=your-key

Directives

Use inline directives in your messages to control TTS per-message:

[[tts:voice=<ref_id>]]     Switch voice
[[tts:speed=1.2]]          Prosody speed (0.5–2.0)
[[tts:model=s1]]           Model override
[[tts:latency=low]]        Latency mode
[[tts:temperature=0.7]]    Sampling temperature
[[tts:top_p=0.8]]          Top-p sampling

Voice Listing

The plugin dynamically lists available voices via /tts voices:

• Official Fish Audio voices (~38 voices)
• Your own cloned/trained voices (marked with "(mine)")

Output Format

The plugin automatically selects the best format based on the channel:

• Voice note channels (Telegram, WhatsApp, Matrix, Feishu) → Opus
• All other channels → MP3

Both formats set voiceCompatible: true — Fish Audio output works cleanly as native voice notes.

Requirements

• OpenClaw ≥ 2026.3.0
• Fish Audio API key (get one here)

License

MIT