OpenClaw already has a nice speech-provider architecture behind messages.tts and supports providers like ElevenLabs, OpenAI, Microsoft, and MiniMax. It would be great to add Gemini 3.1 Flash TTS as a first-class speech provider.

Why this would be valuable

Gemini 3.1 Flash TTS appears to support expressive inline prompt tags / stage directions such as:

• [excitedly]
• [whispers]
• character/style directions
• pacing/delivery cues

That makes it especially appealing for:

• audio-note style replies
• more natural assistant voice messages
• storytelling / dramatic reads
• richer spoken UX

Request

Please add a built-in Gemini speech provider to the messages.tts surface, similar to the existing providers.

Ideally this would support:

1. Provider registration

   • expose Gemini as a normal speech provider under messages.tts.providers.<id>

2. Tag passthrough

   • preserve expressive inline tags / stage directions in TTS input instead of stripping or normalizing them away
   • especially when using [[tts:text]]...[[/tts:text]]

3. Normal OpenClaw integration

   • work with /tts
   • work with auto-TTS flows
   • support normal voice-note/audio attachment delivery
   • support provider selection/fallback like other TTS providers

4. Config shape
   Something roughly like:

   {
     messages: {
       tts: {
         provider: "google",
         providers: {
           google: {
             model: "gemini-3.1-flash-tts",
             voice: "..."
           }
         }
       }
     }
   }

5. Model-driven overrides

   • if possible, let [[tts:text]] and existing TTS directive flows feed expressive tagged text directly into Gemini TTS

Context

From the docs/code, OpenClaw already has a shared speech-provider registry architecture, so this seems like a natural extension rather than a one-off feature.