Gemini 3.5 Transcribe Live

{
  "setup": {
    "generationConfig": {
      "responseModalities": [
        "TEXT"
      ]
    },
    "inputAudioTranscription": {}
  }
}

// Node.js 22+. speech.pcm is 16 kHz, 16-bit, mono PCM; make one with:
// ffmpeg -i speech.mp3 -ar 16000 -ac 1 -f s16le speech.pcm
import { readFile } from "node:fs/promises";

const socket = new WebSocket("wss://api.ohmygpt.com/v1/gemini/live?model=google/gemini-3.5-transcribe-live", ["realtime", "openai-insecure-api-key." + process.env.OHMYGPT_API_KEY]);
socket.onopen = () => {
  socket.send(JSON.stringify({
    setup: {
      generationConfig: { responseModalities: ["TEXT"] },
      inputAudioTranscription: {},
    },
  }));
};
socket.onmessage = async (event) => {
  const data = JSON.parse(typeof event.data === "string" ? event.data : await event.data.text());
  if (data.serverContent?.inputTranscription) console.log(data.serverContent.inputTranscription.text);
  if (!data.setupComplete) return;
  const audio = await readFile("speech.pcm");
  // Send at real-time speed, 100 ms per chunk; the model ends each sentence at a pause
  for (let offset = 0; offset < audio.length; offset += 3200) {
    socket.send(JSON.stringify({ realtimeInput: { audio: { data: audio.subarray(offset, offset + 3200).toString("base64"), mimeType: "audio/pcm;rate=16000" } } }));
    await new Promise((resolve) => setTimeout(resolve, 100));
  }
  socket.send(JSON.stringify({ realtimeInput: { audioStreamEnd: true } }));
  setTimeout(() => socket.close(), 3000);
};
socket.onclose = (event) => event.reason && console.error(event.reason);