Troubleshooting

July 31, 2026 · View on GitHub

Symptoms are grouped by stage: connection, audio capture, and the reply. Open the browser developer console (F12) first. src/index.js logs each stage there.

Connection

The browser refuses to connect, or a WebSocket error appears in the console

Almost always the TLS / mixed-content rule. Browsers block a plain ws:// WebSocket from an https:// page, and only allow ws:// at all to 127.0.0.1.

  • From the hosted (https://) demo, the hub must be reachable over wss:// (TLS). A plain-ws:// hub cannot be used.
  • For local testing, serve the page over http://localhost and connect to a hub on ws://127.0.0.1.

See Configuration: the TLS rule.

"Hivemind connection lost..."

The WebSocket dropped after connecting. Check:

  • The hub is still running and reachable at the IP/port you entered.
  • The Access Key and Password are correct, and the key has not been revoked on the hub (hivemind-core client management). A wrong password fails the V1 handshake, so the connection drops before any audio is sent.
  • Firewall or NAT between you and the hub.

Connect succeeds but nothing happens when you speak

The connection is fine, but the hub is dropping the audio message. Confirm on the hub:

hivemind-core allow-msg "recognizer_loop:b64_audio"

and that the listener is ovos-dinkum-listener >= 0.0.3a19. Without both, audio arrives and is discarded.

Audio capture / VAD

The "Start VAD" button never enables (stays loading)

The VAD failed to initialize. Causes:

  • No network on first load. The VAD model and onnxruntime-web load from a CDN. Check the console for failed CDN requests.
  • Microphone permission denied. Grant it and reload.
  • Insecure context. getUserMedia only works over https:// or http://localhost. Opened any other way, the mic is unavailable.

No microphone permission prompt appears

The page is not in a secure context. Serve it over https:// or http://localhost.

VAD runs but utterances are never sent

Open the console. You should see Speech start / Speech end logs when you talk. If they do not appear, the VAD is not hearing you. Check the OS/browser microphone selection and input level. If they do appear but no reply comes, the problem is on the hub side (see "Connect succeeds but nothing happens" above).

Audio is cut off or split oddly

The VAD decides utterance boundaries. Very short pauses can split a sentence into two sends, and background noise can trigger spurious sends. Each send is listed with a playback control. Play it back to hear exactly what was captured.

Reply

I speak, the audio is listed, but no "HiveMind says:" text appears

The audio reached the page's list (so capture and encode worked), but no speak message came back. This is a hub-side issue:

  • The audio message may be blocked. Re-check allow-msg.
  • STT on the hub may have produced no transcript (silence, noise, wrong language).
  • The intent may not have matched any skill, so nothing spoke.

Inspect the hub logs to see whether the utterance was transcribed and which skill, if any, handled it.

Still stuck

  • Confirm the hosted demo works against a known-good wss:// hub first, to isolate whether the problem is the page or your hub.
  • Cross-check the hub setup against hivemind-mic-satellite. It sends the same recognizer_loop:b64_audio message and has the same hub requirements, so a working mic-satellite confirms the hub side is correct.

← Audio pipeline · Home