import { transcribeAudio } from '../../utils/chatTranscribe'

/**
 * Synchronous speech-to-text for mobile memo fields.
 *
 * POST /api/mobile/transcribe-text
 * Body: multipart form — 'audio' file (webm/ogg/mp3/m4a/wav/aac/mp4), optional 'lang' (de|en|es, default auto)
 * Returns: { status: 200, transcript } — the caller shows/edits the text BEFORE it is
 * persisted anywhere. This differs from /api/mobile/transcribe, which queues the job and
 * appends "[VOICE: …]" to an existing R_Request asynchronously.
 *
 * Fail-soft: when whisper.cpp/ffmpeg are unavailable (dev Mac) the transcript is '' with
 * status 500 so the UI can fall back to typed text.
 */
const MAX_BYTES = 25 * 1024 * 1024

export default defineEventHandler(async (event) => {
  const formData = await readMultipartFormData(event)
  if (!formData) {
    return { status: 400, transcript: '', message: 'Multipart form data required' }
  }

  let audioBuffer: Buffer | null = null
  let audioExt = 'webm'
  let lang: string | undefined

  for (const part of formData) {
    if (part.name === 'audio') {
      audioBuffer = part.data
      // Never trust the client filename — only take a whitelisted extension from it.
      const extMatch = /\.(webm|ogg|oga|mp3|m4a|wav|aac|mp4)$/i.exec(part.filename || '')
      audioExt = extMatch ? extMatch[1].toLowerCase() : 'webm'
    } else if (part.name === 'lang') {
      const l = part.data.toString().trim().toLowerCase().slice(0, 2)
      if (['de', 'en', 'es'].includes(l)) lang = l
    }
  }

  if (!audioBuffer || audioBuffer.length === 0) {
    return { status: 400, transcript: '', message: 'Audio file required' }
  }
  if (audioBuffer.length > MAX_BYTES) {
    return { status: 413, transcript: '', message: 'Audio too large (max 25 MB)' }
  }

  const transcript = await transcribeAudio(audioBuffer, `memo.${audioExt}`, lang)
  if (!transcript) {
    return { status: 500, transcript: '', message: 'Transkription nicht verfügbar' }
  }
  return { status: 200, transcript }
})