Hume AI Not Working? Fix Common Errors

EVI WebSocket not connecting

Hume EVI uses a persistent WebSocket for real-time voice conversation. Correct setup:

// JavaScript — connect to EVI
const socket = new WebSocket(
  `wss://api.hume.ai/v0/evi/chat?api_key=YOUR_HUME_API_KEY`
);

socket.onopen = () => {
  // Send initial session settings before any audio
  socket.send(JSON.stringify({
    type: "session_settings",
    audio: { encoding: "linear16", sample_rate: 16000, channels: 1 }
  }));
};

socket.onmessage = (event) => {
  const msg = JSON.parse(event.data);
  if (msg.type === "audio_output") {
    // Play base64-encoded audio response
  }
};

• The API key goes in the query string, not a header
• Wait for socket.onopen before sending audio
• Corporate proxies blocking wss:// will cause silent failures

Expression Measurement API — 401 authentication

The Expression Measurement API uses a different header than most AI APIs:

import requests

response = requests.post(
    "https://api.hume.ai/v0/batch/jobs",
    headers={"X-Hume-Api-Key": "YOUR_HUME_API_KEY"},
    json={
        "models": {"face": {}},
        "urls": ["https://example.com/video.mp4"]
    }
)
print(response.json())

Browser audio not working (microphone blocked)

EVI requires microphone access via the browser's getUserMedia API. Common blockers:

• HTTP vs HTTPS: getUserMedia is blocked on insecure origins. Use HTTPS or localhost.
• Permission denied: The user dismissed or blocked the permission dialog. Check browser site settings and request permission again.
• No microphone device: Test with navigator.mediaDevices.enumerateDevices() to list available audio inputs.
• Sending audio before session open: Always wait for the WebSocket onopen event before streaming audio.

High latency / slow responses

EVI latency has several components. The part you can control: