Transcription API

{
  "provider": "soniox",
  "soniox": {
    "context": "{\"text\": \"Customer support calls for Breeze Buddy\", \"terms\": [\"Breeze Buddy\", \"UPI autopay\"], \"general\": [{\"key\": \"domain\", \"value\": \"payments\"}]}"
  }
}

# 1. Get a temporary JWT from the login API (server-side only — the
#    email/password are credentials and must never reach client code)
TOKEN=$(curl -s -X POST $BASE_URL/agent/voice/breeze-buddy/login \
  -H "Content-Type: application/json" \
  -d '{"email": "[email protected]", "password": "..."}' | jq -r .access_token)
# 2. Transcribe a clip (Soniox with domain context — a pinned, strict request)
curl -X POST $BASE_URL/agent/voice/breeze-buddy/stt/transcribe \
  -H "Authorization: Bearer $TOKEN" \
  -F "[email protected]" \
  -F "provider=soniox" \
  -F 'soniox={"context": "Breeze Buddy, Juspay, UPI autopay"}'

{
  "text": "Hi, I'm calling about the UPI autopay setup on my Breeze Buddy account.",
  "provider": "soniox",
  "model": "stt-async-v2"
}

{
  "provider": "deepgram",
  "model": "nova-3-general",
  "language": "en",
  "sample_rate": 16000,
  "deepgram": { "smart_format": true, "endpointing_ms": 25 }
}

// Your server is the credential proxy: email/password live in env vars
// here and NEVER ship to the browser. Gate this route with your own
// product's user auth, then hand out only the short-lived JWT.
app.get('/api/stt-token', requireYourAppAuth, async (req, res) => {
  const login = await fetch(BREEZE_BASE + '/agent/voice/breeze-buddy/login', {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      email: process.env.BREEZE_EMAIL,
      password: process.env.BREEZE_PASSWORD
    })
  }).then((r) => r.json());
  res.json({ access_token: login.access_token });
});

// Browser: fetch the temporary token from YOUR backend — the browser
// never sees Breeze Buddy credentials and never calls /login itself.
const { access_token } = await fetch('/api/stt-token').then((r) => r.json());
// Connect the stream — browsers pass the token as a query param
const ws = new WebSocket(
  WS_BASE + '/agent/voice/breeze-buddy/stt/stream?token=' + access_token
);
ws.binaryType = 'arraybuffer';
ws.onopen = () =>
  ws.send(JSON.stringify({ provider: 'soniox', sample_rate: 16000, language: 'en' }));
ws.onmessage = (msg) => {
  const event = JSON.parse(msg.data);
  if (event.type === 'ready') startMic();
  if (event.type === 'partial') renderInterim(event.text);
  if (event.type === 'final') appendLine(event.text);
  if (event.type === 'error') console.error(event.message);
};
// Mic -> PCM16 mono at 16 kHz -> binary frames
async function startMic() {
  const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
  const ctx = new AudioContext({ sampleRate: 16000 });
  const source = ctx.createMediaStreamSource(stream);
  const tap = ctx.createScriptProcessor(4096, 1, 1); // use an AudioWorklet in production
  tap.onaudioprocess = (e) => {
    const f32 = e.inputBuffer.getChannelData(0);
    const pcm = new Int16Array(f32.length);
    for (let i = 0; i < f32.length; i++) {
      const s = Math.max(-1, Math.min(1, f32[i]));
      pcm[i] = s < 0 ? s * 0x8000 : s * 0x7fff;
    }
    if (ws.readyState === WebSocket.OPEN) ws.send(pcm.buffer);
  };
  source.connect(tap);
  tap.connect(ctx.destination);
}
// End the session: server flushes trailing finals, then closes with 1000
function stop() {
  ws.send(JSON.stringify({ type: 'stop' }));
}