Voice

Voice is a first-class input and output mode in Crew. The default provider — Piper — runs locally, so nothing leaves your machine. Cloud STT via AWS Transcribe is an optional extra.

• Speech-to-text (STT) — press the mic button and start talking; live partials stream into the chat input
• Text-to-speech (TTS) — replies are spoken aloud through your speakers, streamed word-by-word as the agent generates

Both are on-demand. Voice input never fires until you press the mic; voice output never plays unless you enable it per-session or globally.

Hold or click the mic button in the dashboard chat input.

• Live partials appear in the textarea as you speak
• Silence for ~1.5s auto-finalizes
• Click the button again to stop early

Behind the scenes, this streams audio to /api/stt (WebSocket), which pipes it to your chosen provider and returns transcription events in real time.

Configure the STT provider in ~/.kiro/crew/config.json:

{
  "voice": {
    "stt_provider": "whisper"
  }
}

Install the cloud STT extra:

pip install "kirocrew[voice]"

Attach a voice memo as a file with @filename — the audio is transcribed automatically and inlined into the message. Useful when you want to leave a voice note in an existing message without triggering the live mic.

Piper is a fast, local neural TTS engine. It runs entirely on your machine (no internet needed) and streams audio word-by-word as the agent generates its reply.

• Enable TTS from the chat tab header (🔊 icon)
• The reply is spoken as it streams — you hear the first sentence before the model finishes the last
• Toggle off with the same button

Piper voice files are downloaded on first use, then cached locally.

{
  "voice": {
    "tts_provider": "piper",
    "tts_voice": "en_US-amy-medium",
    "tts_speed": 1.0
  }
}

Available voices: any Piper voice file. Drop a .onnx file into ~/.kiro/crew/voices/ and reference it by filename.

TTS is designed to fade into the background:

• Code blocks are skipped (they'd be unlistenable)
• Tool call blocks are skipped
• Long URLs are read as "URL" instead of character-by-character
• Markdown headings are read with a slight pitch shift

You can override any of these from the voice settings panel.

Voice output can be interrupted the same way as a chat turn:

• Click ⏹ Stop to cancel both the streaming reply and the TTS
• The !stop command in Slack does the same for cross-surface use

The interrupt is bidirectional: stopping the reply also stops the TTS mid-word.

Voice input is dashboard-only for now. Slack voice memos work through Slack's voice-note feature. There's no voice mode in the CLI.

Voice output works everywhere the dashboard is open — including remote hosts accessed over an SSH tunnel or mobile tunnel. The audio stream is delivered as chunked WebSocket messages, so latency depends on your network.

• Piper TTS — fully local; no audio leaves the machine
• Whisper STT — local when installed; falls back to browser STT (which may use a browser provider's cloud service)
• AWS Transcribe — audio is streamed to AWS; opt in only with the voice extra installed and credentials configured

Piper is the default because it's the only provider that never leaves the machine.