pediatric-ai-scribe-v3/src/routes/transcribe.js

const express = require('express');
const router = express.Router();
const multer = require('multer');
const { whisperClient, litellmClient } = require('../utils/ai');
const { transcribeWithAWS, isAWSTranscribeConfigured } = require('../utils/transcribeAWS');
const { transcribeWithLocal, isLocalWhisperConfigured } = require('../utils/transcribeLocal');
const { authMiddleware } = require('../middleware/auth');

const upload = multer({ storage: multer.memoryStorage(), limits: { fileSize: 25 * 1024 * 1024 } });

// Determine which provider to use:
// TRANSCRIBE_PROVIDER=litellm → LiteLLM proxy (any Whisper-compatible model)
// TRANSCRIBE_PROVIDER=aws     → Amazon Transcribe
// TRANSCRIBE_PROVIDER=local   → local whisper.cpp / faster-whisper
// TRANSCRIBE_PROVIDER=openai  → OpenAI Whisper directly
// (auto) LITELLM_API_BASE set → litellm
// (auto) AWS Bedrock region set → aws
// (default) → openai whisper
function getTranscribeProvider() {
  var env = process.env.TRANSCRIBE_PROVIDER;
  if (env === 'litellm') return 'litellm';
  if (env === 'local')   return 'local';
  if (env === 'aws')     return 'aws';
  if (env === 'openai')  return 'openai';
  // Auto-detect: only use LiteLLM for STT if LITELLM_STT_MODEL is explicitly set
  // Having LITELLM_API_BASE for AI text does NOT automatically route audio through LiteLLM
  if (process.env.LITELLM_STT_MODEL && process.env.LITELLM_API_BASE && litellmClient) return 'litellm';
  if (isAWSTranscribeConfigured()) return 'aws';
  return 'openai';
}

function isTranscribeAvailable() {
  if (isLocalWhisperConfigured()) return true;
  if (isAWSTranscribeConfigured()) return true;
  if (litellmClient) return true;
  if (whisperClient) return true;
  return false;
}

var provider = getTranscribeProvider();
var medical = process.env.AWS_TRANSCRIBE_MEDICAL === 'true';
var available = isTranscribeAvailable();
console.log('🎙️  Transcribe provider:', provider + (provider === 'aws' && medical ? ' (Medical)' : '') + (available ? '' : ' (NOT CONFIGURED — browser speech only)'));

// Status endpoint — frontend checks this to decide whether to upload audio
router.get('/transcribe/status', authMiddleware, (req, res) => {
  res.json({ available: available, provider: available ? provider : 'none' });
});

router.post('/transcribe', authMiddleware, upload.single('audio'), async (req, res) => {
  try {
    if (!req.file) return res.status(400).json({ error: 'No audio' });
    var startTime = Date.now();
    var fileSize = req.file.size;
    console.log('[Transcribe] Received ' + (fileSize / 1024).toFixed(0) + 'KB audio (' + (req.file.mimetype || 'unknown') + ') via ' + provider);

    if (provider === 'local') {
      var text = await transcribeWithLocal(req.file.buffer, req.file.mimetype || 'audio/webm');
      console.log('[Transcribe] Local done in ' + (Date.now() - startTime) + 'ms');
      return res.json({ success: true, text: text, provider: 'local-whisper', duration: Date.now() - startTime });
    }

    if (provider === 'aws') {
      if (!isAWSTranscribeConfigured()) {
        return res.status(400).json({ error: 'AWS Transcribe not configured. Set AWS_BEDROCK_REGION.' });
      }
      var text = await transcribeWithAWS(req.file.buffer, req.file.mimetype || 'audio/webm');
      console.log('[Transcribe] AWS done in ' + (Date.now() - startTime) + 'ms');
      return res.json({ success: true, text: text, provider: 'aws-transcribe', duration: Date.now() - startTime });
    }

    if (provider === 'litellm') {
      if (!process.env.LITELLM_API_BASE) return res.status(400).json({ error: 'LITELLM_API_BASE not set.' });
      var sttModel = process.env.LITELLM_STT_MODEL || 'whisper-1';
      var FormData = require('form-data');
      var axios = require('axios');
      var form = new FormData();
      form.append('file', req.file.buffer, {
        filename: 'audio.webm',
        contentType: req.file.mimetype || 'audio/webm'
      });
      form.append('model', sttModel);
      form.append('language', 'en');
      form.append('response_format', 'json');
      form.append('prompt', 'Medical patient encounter. Pediatric. Clinical terms, diagnoses, medications.');
      var sttHeaders = Object.assign({}, form.getHeaders());
      if (process.env.LITELLM_API_KEY) sttHeaders['Authorization'] = 'Bearer ' + process.env.LITELLM_API_KEY;
      var base = process.env.LITELLM_API_BASE.replace(/\/+$/, '');
      var sttResp = await axios.post(base + '/v1/audio/transcriptions', form, {
        headers: sttHeaders,
        timeout: 120000
      });
      var text = (sttResp.data && sttResp.data.text) ? sttResp.data.text : String(sttResp.data || '');
      console.log('[Transcribe] LiteLLM (' + sttModel + ') done in ' + (Date.now() - startTime) + 'ms');
      return res.json({ success: true, text: text, provider: 'litellm-stt', duration: Date.now() - startTime });
    }

    // OpenAI Whisper (direct)
    if (!whisperClient) return res.status(400).json({ error: 'Whisper not configured. Set OPENAI_API_KEY.' });
    var file = new File([req.file.buffer], 'audio.webm', { type: req.file.mimetype || 'audio/webm' });
    var result = await whisperClient.audio.transcriptions.create({
      file, model: 'whisper-1', language: 'en',
      response_format: 'text',
      prompt: 'Medical patient encounter. Pediatric. Clinical terms, diagnoses, medications.'
    });
    var text = typeof result === 'string' ? result : result.text;
    console.log('[Transcribe] Whisper done in ' + (Date.now() - startTime) + 'ms');
    res.json({ success: true, text: text, provider: 'openai-whisper', duration: Date.now() - startTime });
  } catch (err) {
    console.error('[Transcribe] Error:', err.message);
    res.status(500).json({ error: err.message });
  }
});

module.exports = router;