That old iPhone is a $0 streaming computer.
Plug a hydrophone into it.

Why iPhone, not Android, not a Pi

Hardware path

   ┌──────── water ────────┐    ┌──────────────── deck ────────────────┐
   │                       │    │                                       │
   │  Aquarian H2a-XLR     │    │   XLR-F  ─►  Mogan-style XLR-to-      │
   │  (10Hz–100kHz, ±1dB)  │────┼──────►       Lightning interface       │
   │  20 m cable           │    │              (or Apogee MiC Plus       │
   │                       │    │               via USB-C lightning)     │
   └───────────────────────┘    │                            │           │
                                │                            ▼           │
                                │                ┌─────────────────────┐ │
                                │                │  iPhone 8 / SE 2    │ │
                                │                │  iOS 16+            │ │
                                │                │  Larix or PWA       │ │
                                │                │  CoreML orca model  │ │
                                │                │  H.264 + AAC LC     │ │
                                │                └──────────┬──────────┘ │
                                │                           │            │
                                │                           ▼ RTSP/SRT   │
                                │                  ┌─────────────────┐   │
                                │                  │  ingest server  │   │
                                │                  │  (MediaMTX)     │   │
                                │                  └────────┬────────┘   │
                                │                           ▼            │
                                │           HLS public + WebSocket meta  │
                                │           → salish-sigint dashboard          │
                                └──────────────────────────────────────-─┘

Bill of materials

Software stack — the easy way

Capture & encode: Larix Broadcaster (free)

Larix has been the de-facto iOS RTMP/RTSP/SRT pusher for a decade. Audio-only mode is one toggle. The settings that matter for hydrophone work:

Sample rate

48 kHz (96 kHz if your interface supports it — orca clicks reach 80 kHz)

Channels

Mono

Codec

AAC-LC at 256 kbps for archive · Opus 64 kbps for live preview

Connection

SRT to your ingest, with latency=2000 ms buffer for cellular

Auto-restart

ON — phone will drop and resume gracefully

Background audio

ON via Settings → Background Modes → Audio

Ingest: MediaMTX on a $5 VPS

# mediamtx.yml — ingest SRT, expose HLS + RTSP + WebRTC simultaneously
paths:
  hydrophone-pier70:
    source: publisher
    publishUser: bay-station
    publishPass: ${HYDROPHONE_KEY}

srt:
  enable: yes
  listen: :8890

hls:
  enable: yes
  variant: lowLatency
  segmentDuration: 1s

webrtc:
  enable: yes
  iceServers: [stun:stun.l.google.com:19302]

Spin up on a $5 Hetzner / Vultr box, point Larix at srt://your.host:8890?streamid=publish:hydrophone-pier70, done. HLS playback URL is https://your.host/hydrophone-pier70/index.m3u8 — paste into the Salish SIGINT dashboard.

Software stack — the no-install way (PWA)

If you don't want to install anything, modern Safari supports getUserMedia({audio: true}) + WebRTC + MediaRecorder. You can run this exact page from the iPhone, hit start, and stream. Tradeoff: less robust to backgrounding than Larix, and Safari does cap session length without a Lock-Screen-Audio shim.

// Run from iPhone Safari — works as a streaming source w/ no install
async function startHydrophone() {
  const stream = await navigator.mediaDevices.getUserMedia({
    audio: {
      sampleRate: 48000,
      channelCount: 1,
      echoCancellation: false,    // off, off, off
      autoGainControl: false,
      noiseSuppression: false,
    }
  });

  // Push to a media server with WHIP (WebRTC HTTP Ingestion Protocol)
  const pc = new RTCPeerConnection();
  stream.getTracks().forEach(t => pc.addTrack(t, stream));
  const offer = await pc.createOffer();
  await pc.setLocalDescription(offer);

  const r = await fetch('https://your.host/whip/hydrophone-pier70', {
    method: 'POST',
    headers: { 'Content-Type': 'application/sdp' },
    body: offer.sdp,
  });
  await pc.setRemoteDescription({ type: 'answer', sdp: await r.text() });
}

On-device detection — CoreML

Run a YAMNet backbone (Google's 521-class audio classifier) finetuned on the OrcaSound corpus. Tiny model, trivial to convert to CoreML with coremltools. The phone does VAD-style analysis on a 1-s rolling window and pushes a metadata event on detection. The audio keeps streaming regardless; the metadata just lets the dashboard mark interesting moments.

# Train: PyTorch / TF, export to CoreML
import coremltools as ct
mlmodel = ct.convert(
    yamnet_finetuned,
    inputs=[ct.TensorType(shape=(1, 15600))],   # 0.975 s @ 16 kHz
    classifier_config=ct.ClassifierConfig(["orca-call", "vessel", "other"]),
    compute_units=ct.ComputeUnit.CPU_AND_NE,         # Neural Engine
)
mlmodel.save("OrcaDetector.mlpackage")

// Swift — push detection events as RTSP onMetadata or a sidecar WebSocket
let model = try OrcaDetector(configuration: .init())
let result = try model.prediction(audio: window)
if result.classLabel == "orca-call", result.classProbability["orca-call"]! > 0.7 {
  meta.send([
    "t": Date().timeIntervalSince1970,
    "event": "orca-call",
    "conf": result.classProbability["orca-call"]!,
  ])
}

Mounting & the boring stuff that breaks first

• Cable strain relief. The hydrophone cable will whip in current. Loop it once around the dock cleat with a soft tie before the gland.
• Phone heat. A Pelican in direct sun hits 60 °C and the phone will throttle. Stick it on the shaded side of a piling, or use a small white solar-shed.
• Power. A 20,000 mAh USB battery runs Larix audio-only for ~36 h. Real deployments want shore power or a Voltaic V25/V44 panel pair.
• Wi-Fi vs. cellular. Marina Wi-Fi is rate-limited. A data-only SIM ($15/mo) is almost always the right answer. Test in the spot before you commit.
• Two phones is one phone. Run a backup. iPhones lock up; the second phone is your redundancy.

Public hydrophone networks worth comparing yourself to

• OrcaSound — Bush Point, Port Townsend, Lime Kiln, Orcas Island. Open audio archive going back years. Use it as ground-truth pre-deployment.
• Ocean Observatories Initiative cabled array — NOAA-funded broadband hydrophones off Oregon. Public data, calibrated.
• Whale.fm (offline now, archives still up) — citizen-science classification, useful as a UX reference.

Where it fits in the stack