CPU-first speech-to-text.

// Live audio → partial + final transcripts
const ws = new WebSocket(
  "wss://app.kroko.ai/api/v1/transcripts/streaming" +
  "?apiKey=YOUR_KEY&languageCode=en-US"
);

ws.onmessage = (e) => {
  const r = JSON.parse(e.data);
  if (r.type === "partial") showLive(r.text);
  if (r.type === "final")   appendSegment(r.text);
};
// then send 16 kHz mono PCM chunks

// Recorded file → final transcript + timestamps
const ws = new WebSocket(
  "wss://app.kroko.ai/api/v1/transcripts/pre-recorded" +
  "?apiKey=YOUR_KEY&languageCode=en-US"
);

// send metadata header + audio chunks,
// then "Done" when the upload completes
ws.onmessage = (e) => {
  const { segments } = JSON.parse(e.data);
  render(segments); // word + segment timing
};

# Run the same models on your own box, no GPU
docker run -p 8080:8080 kroko/asr:on-prem

# Point the WebSocket client at your server
wss://localhost:8080/api/v1/transcripts/streaming

# Streaming + post-processed models, fully private