Talking to Wally — voice input and read-aloud responses

Wally has three voice features — say "Hey Wally" hands-free to dictate, click the mic to dictate manually, and have Wally's reply read aloud. All three are signed-in only and billed to your org as part of pay-as-you-go usage.

Hands-free wake word ("Hey Wally"):
• When the Wally drawer is open, Wally is quietly listening for the phrase "Hey Wally"
• Once you say it, Wally starts recording your message and the listening row turns into a red "Recording…" indicator
• Stop talking for ~1.8 seconds and recording auto-stops; the transcript is sent to Wally without you needing to click anything
• Variants like "Hi Wally" or "OK Wally" also work
• To turn it off, click the mic-pulse icon in the drawer header (it shows a slash when off) — the choice persists until you turn it back on
• Hands-free needs Chrome, Edge, or Safari (Web Speech API). On Firefox the toggle is hidden and only the manual mic is available

Voice input (manual mic):
• Click the microphone icon to the left of the message box
• A pulsing red dot and a running timer show that recording is live
• Recording stops automatically after ~1.5 seconds of silence — you can also tap the mic again to stop immediately
• The transcript drops into the input box for you to review and edit before sending (so you can correct mishears before it goes to Wally)
• If you've already typed something, the transcript is appended on the end
• If your browser denies mic access, allow it from your browser's site settings and try again

Read aloud (speaker icon):
• Each Wally response shows a small speaker icon when you hover it; click to play it back through your system audio
• Click again on the same message to stop playback
• Audio is generated by ElevenLabs and cached in your private Wally storage, so replaying a reply doesn't re-bill

Auto-speak:
Account → Preferences → Wally Voice → toggle "Auto-speak responses" — every new Wally reply starts playing automatically once the assistant finishes streaming. Pair it with hands-free for a fully voice-driven loop.

Turning features off:
Same Preferences panel — toggle "Voice input" off to hide the mic, "Read aloud" off to hide the speaker icons, or "Hands-free wake word" off to disable the always-listening behaviour. Auto-speak is its own switch and only fires when read-aloud is on. The hands-free header toggle and the Preferences toggle are the same setting.

Privacy notes:
• The wake-word listener runs entirely in your browser via the Web Speech API — audio for "Hey Wally" detection never leaves your device
• Once a recording is captured, the bytes are sent to Google Cloud Speech-to-Text for transcription, then deleted right after
• Voice memos are uploaded to your private wallyChat storage namespace (gated by storage rules to your user only) and deleted right after transcription
• TTS audio files are cached under wallyChat/{uid}/tts/ and only readable by you
• Wally chat audio bypasses the platform image moderation pipeline — that pipeline only inspects images

Languages:
Google Cloud Speech-to-Text is given your platform language as a hint for transcription accuracy; ElevenLabs picks an appropriate voice model based on the same hint.