OpenAI gpt-transcribe example: Dutch WAV + BHV-Knop terms (no API keys)

transcribe-bhv.ts

/**
 * Example: Dutch speech-to-text with OpenAI `gpt-transcribe`.
 *
 *   export OPENAI_API_KEY=...
 *   npx tsx transcribe-bhv.ts ./clip.wav
 *
 * Do not hardcode API keys. Restricted project keys need audio transcription
 * scope; `whisper-1` may be blocked even when `gpt-transcribe` works.
 */
import { readFile } from "node:fs/promises";
import { basename } from "node:path";

const TRANSCRIBE_URL = "https://api.openai.com/v1/audio/transcriptions";
const MODEL = "gpt-transcribe";

/** Prompt should be in the same language as the audio. */
const PROMPT =
  "Nederlandse opname over BHV-Knop.nl, een compleet BHV-oproepsysteem. " +
  "Vaak genoemde termen: BHV, BHV-er, BHV-ers, BHV-Knop, BHV-app, " +
  "BHV-organisatie, BHV-oproepsysteem, BHV-groepsgesprek, BHV-plan, " +
  "basisstation, aanwezigheidskiosk, kiosk, alarmafhandeling, " +
  "aanwezigheidsdetectie, bedrijfshulpverlening, bedrijfsongeval, " +
  "portofoon, aanmelden, afmelden, centrale post, akoestisch signaal, " +
  "groepsgesprek.";

/** Literal terms that STT otherwise mangles (BHV, knop, kiosk, …). */
const KEYWORDS = [
  "BHV",
  "BHV-Knop",
  "BHV-Knop.nl",
  "BHV-app",
  "BHV-oproepsysteem",
  "BHV-organisatie",
  "BHV-groepsgesprek",
  "BHV-plan",
  "bedrijfshulpverlening",
  "bedrijfsongeval",
  "basisstation",
  "aanwezigheidskiosk",
  "aanwezigheidsdetectie",
  "alarmafhandeling",
  "portofoon",
  "kiosk",
  "aanmelden",
  "afmelden",
  "centrale post",
  "akoestisch signaal",
  "groepsgesprek",
];

type TranscriptionResponse = {
  text: string;
  languages?: { code: string }[];
  usage?: { type: string; seconds?: number };
  error?: { message: string; code?: string };
};

async function transcribe(audioPath: string, apiKey: string): Promise<TranscriptionResponse> {
  const bytes = await readFile(audioPath);
  const form = new FormData();

  form.set("model", MODEL);
  form.set("prompt", PROMPT);
  form.set(
    "file",
    new Blob([bytes], { type: "audio/wav" }),
    basename(audioPath),
  );
  // `gpt-transcribe` uses `languages` (list), not `language`.
  form.append("languages[]", "nl");
  for (const keyword of KEYWORDS) {
    form.append("keywords[]", keyword);
  }

  const response = await fetch(TRANSCRIBE_URL, {
    method: "POST",
    headers: { Authorization: `Bearer ${apiKey}` },
    body: form,
  });

  const body = (await response.json()) as TranscriptionResponse;
  if (!response.ok) {
    throw new Error(body.error?.message ?? `HTTP ${response.status}`);
  }
  return body;
}

const apiKey = process.env.OPENAI_API_KEY;
const audioPath = process.argv[2];

if (!apiKey) {
  throw new Error("Set OPENAI_API_KEY in the environment.");
}
if (!audioPath) {
  throw new Error("Usage: npx tsx transcribe-bhv.ts ./clip.wav");
}

const result = await transcribe(audioPath, apiKey);
console.log(result.text);
if (result.languages?.length) {
  console.error("languages:", result.languages.map((item) => item.code).join(", "));
}