TinyClaw: Implement feature-gated voice transcription pipeline

[AlexMikhalev]

Summary

Implement real voice transcription in TinyClaw with feature-gated enablement and graceful fallback.

Extend Existing Terraphim Functionality

• Reuse Terraphim provider/routing patterns where applicable.
• Keep voice as a composable tool in TinyClaw tool registry.

Implementation Notes

• crates/terraphim_tinyclaw/src/tools/voice_transcribe.rs
• crates/terraphim_tinyclaw/src/config.rs
• crates/terraphim_tinyclaw/Cargo.toml
• voice integration tests

Acceptance Criteria

• voice_transcribe downloads, validates, transcribes, and cleans up temp files.
• voice feature gate controls compilation/runtime behavior.
• Failure modes return useful user-facing messages (provider unavailable, invalid audio, timeout).
• Integration tests validate pipeline and error handling.

References

• docs/plans/tinyclaw-openclaw-parity-design-2026-02-27.md (Step 4)

[AlexMikhalev]

Track 4 Complete: Voice Transcription Pipeline

Implementation Summary

Step 4.1: Add voice dependencies and implement feature-gated transcription - COMPLETE

Changes Made:

1. Cargo.toml: Added voice dependencies behind feature flag:

   • whisper-rs = { version = "0.11", optional = true }
   • symphonia = { version = "0.5", optional = true, features = ["ogg", "pcm", "mp3", "wav"] }
   • hound = { version = "3.5", optional = true }
   • voice = ["dep:whisper-rs", "dep:symphonia", "dep:hound"]

2. voice_transcribe.rs: Complete rewrite with:

   • Feature-gated implementation using #[cfg(feature = "voice")]
   • Audio download from URLs with extension validation (OGG, MP3, WAV, M4A, WebM)
   • Audio format conversion to WAV 16kHz mono using symphonia
   • Whisper transcription using whisper-rs with correct WhisperState API
   • Graceful fallback message when voice feature not enabled
   • Proper error handling with user-facing messages

Build Commands:

# Without voice feature (default)
cargo build -p terraphim_tinyclaw

# With voice feature
cargo build -p terraphim_tinyclaw --features voice

Acceptance Criteria Status:

• voice_transcribe downloads, validates, transcribes, and cleans up temp files
• voice feature gate controls compilation/runtime behavior
• Failure modes return useful user-facing messages (provider unavailable, invalid audio, timeout)
• Tests validate pipeline and error handling (135 tests passing)

Notes:

• The terraphim_ai_nodejs crate has pre-existing napi-rs compilation errors unrelated to these changes
• All tinyclaw-specific tests pass successfully

Co-Authored-By: Claude Sonnet 4.6 noreply@anthropic.com