Holler 📢

The Google Home broadcast app for families.

One tap on a parent's phone → attention chime + spoken message on the Google/Nest speakers you choose, forced loud, interrupting whatever's playing. Call the kids to dinner, get someone out of their room, announce "we're leaving in five" — and actually be heard over headphones-off music and closed doors, where Google's built-in broadcast (fixed, polite volume) gets ignored. One-way by design. LAN-only, self-hosted: no cloud, no accounts, no data leaves the house.

The app	Admin: speaker & volume	Admin: edit messages

[Parent phone PWA] --HTTP--> [FastAPI (Docker)] --pychromecast--> [Google/Nest speaker]
                                  |
                                  +-- serves pre-rendered audio over LAN HTTP
                                  +-- presets.yaml (edited via /admin page)

Per tap: connect to the pinned speaker → stash its current volume → set the broadcast volume → cast the pre-rendered file → wait for playback → restore the stashed volume. Concurrent taps get "already playing" so two parents can't stomp each other's volume capture/restore. Whatever the kid was playing does not auto-resume — feature, not bug.

Setup

1. Get it running

Docker (recommended, e.g. on a home server):

git clone https://github.com/dstahl11/holler.git && cd holler
cp presets.example.yaml presets.yaml
docker compose up -d --build

The compose file uses host networking so the speaker can fetch audio straight from your server's LAN IP. App: http://<server-ip>:8000.

Bare Python (dev / Mac):

python3 -m venv .venv && .venv/bin/pip install -r requirements.txt
cp presets.example.yaml presets.yaml
.venv/bin/uvicorn app.main:app --host 0.0.0.0 --port 8000

Set HOLLER_DRY_RUN=1 to simulate casts while you play with the UI.

2. Point it at your speaker

Open http://<server>:8000/admin (or the ⚙︎ gear in the app):

1. Tap Scan network — every Cast device in the house shows up.
2. Tap your kid's speaker to add it. Done.

Add as many speakers as you like — broadcasts play on every speaker that's switched on in the list (kid's room + hallway, two kids' rooms, …), each with its own volume capture/restore. Toggle them without removing.

Give each speaker a DHCP reservation in your router so its IP never moves. (CLI alternative: python scripts/discover.py, then edit presets.yaml.)

3. Make it say your kid's name

Still in /admin: edit the message text on each card (the examples say "Sam"), then Save changes — audio re-renders automatically. TTS engines:

• macOS say — used automatically when running bare on a Mac. Not available inside Docker (Linux) — Piper takes over there.
• Piper (self-hosted neural TTS) — bundled in the Docker image with the en_US-lessac-medium voice, works out of the box. Audio rendered earlier on a Mac keeps working after a move to Docker — they're just WAV files; only future wording changes render with Piper.
• Recorded parent voice — the one kids actually respond to. Record each phrase on your phone, drop the files at audio/raw/<preset-id>.m4a (any format), Save/re-render. Recordings always win over TTS.

Every rendered file is chime + phrase baked into one normalized WAV, ~3–8s.

4. ⚠️ Run the loudness test

Tap the 🔔 Test button (chime only) with the kid's door closed and their own audio playing. Max loudness varies a lot by model — a Nest Mini caps much lower than a Nest Audio or full Home. If it doesn't cut through: cast to a speaker group, move a bigger speaker into the room, or add one in the hallway. Volume slider lives in /admin (field-tested: 100% on a full Google Home produced a formal complaint from the target; 70% is plenty).

5. Install on the parents' phones

Open the app URL in Safari (iOS) or Chrome (Android) → Share → Add to Home Screen. Opens full-screen with big buttons. That's the entire daily UX: open, tap, done.

Browsers only enable service workers over HTTPS, so over plain LAN HTTP this is a home-screen shortcut rather than a fully installed PWA — which is all it needs. Add Tailscale + HTTPS later and offline caching lights up by itself.

Administration

Everything lives in /admin — scan/pick speaker, add/reword/delete messages (Save re-renders audio), broadcast volume, and two independent PINs:

• App PIN — gates the button grid, if the kids discover the URL and start counter-broadcasting.
• Admin PIN — gates this settings page.

Changes hot-reload; no restart. Saves write presets.yaml (previous version kept as presets.yaml.bak), so hand-editing the YAML still works too.

API

• GET /api/presets — button list for the PWA
• POST /api/broadcast/{id} — do it (X-Pin header if PIN set). 200 sent, 401 bad PIN, 429 already playing, 502 speaker unreachable
• GET /api/health — is the speaker reachable
• GET|PUT /api/admin/config, POST /api/admin/scan, POST /api/admin/render — the admin page's API (X-Admin-Pin header if set)

Config reference (presets.yaml)

Key	Meaning
devices[]	name, host, optional uuid, enabled — broadcasts hit every enabled one
broadcast.volume	0.0–1.0 forced during broadcast
broadcast.play_timeout	Max seconds to wait before restoring volume
broadcast.advertise_host	LAN IP the speaker fetches audio from; auto-detected if empty
security.pin / security.admin_pin	App / admin PIN gates; "" disables
tts.engine	auto / say / piper — auto prefers audio/raw/ recordings
presets[]	id, label, emoji, tts (or file: for custom audio)

Design notes

• Cast, not Google's native Broadcast — native Broadcast plays at a fixed volume you can't override; casting lets us force loud. Loud won.
• One-way on purpose — the only "reply" path Google offers conflicts with forced volume, and we're optimizing for impossible to ignore.
• Pre-rendered audio, not TTS at trigger time — lower latency, predictable output, no runtime dependency.
• LAN-only — bind it to your home network and don't port-forward it. The attack surface is "people on your Wi-Fi," and there's a PIN for them.