docs: document voice plugin commands

Add Voice STT/TTS section covering !listen, !say, and optional
[voice] config block with whisper_url, piper_url, silence_gap.

docs/USAGE.md

@@ -1677,3 +1677,40 @@ file (natural dedup).
 - Use `!kept` to list preserved files and their sizes
 - Use `!kept clear` to delete all preserved files
 - On cancel/error, files are not deleted (needed for `!resume`)
+
+### Voice STT/TTS
+
+Transcribe voice from Mumble users via Whisper STT and speak text aloud
+via Piper TTS. Requires local Whisper and Piper services.
+
+```
+!listen [on|off]     Toggle voice-to-text transcription (admin)
+!listen              Show current listen status
+!say <text>          Speak text aloud via TTS (max 500 chars)
+```
+
+STT behavior:
+
+- When enabled, the bot buffers incoming voice PCM per user
+- After a configurable silence gap (default 1.5s), the buffer is
+  transcribed via Whisper and posted as an action message
+- Utterances shorter than 0.5s are discarded (noise filter)
+- Utterances are capped at 30s to bound memory and latency
+- Transcription results are posted as: `* derp heard Alice say: hello`
+- The listener survives reconnects when `!listen` is on
+
+TTS behavior:
+
+- `!say` fetches WAV from Piper and plays it via `stream_audio()`
+- Piper outputs 22050Hz WAV; ffmpeg resamples to 48kHz automatically
+- TTS shares the audio output with music playback
+- Text is limited to 500 characters
+
+Configuration (optional):
+
+```toml
+[voice]
+whisper_url = "http://192.168.122.1:8080/inference"
+piper_url = "http://192.168.122.1:5000/"
+silence_gap = 1.5
+```