v83: обрезка 0.2с на сервере (ffmpeg в прокси) — надёжнее браузерного WAV

server/proxy.py

@@ -1,6 +1,9 @@
 #!/usr/bin/env python3
 from flask import Flask, request, Response
 import requests
+import subprocess
+import tempfile
+import os
 
 app = Flask(__name__)
 GROQ_BASE = "https://api.groq.com"
@@ -15,13 +18,17 @@ CORS = {
 def options(path):
     return Response(status=204, headers=CORS)
 
-# Generic proxy for ALL Groq paths
 @app.route("/<path:subpath>", methods=["GET", "POST", "PUT", "DELETE", "PATCH"])
 def proxy(subpath):
     url = f"{GROQ_BASE}/{subpath}"
     hdrs = {}
     if request.headers.get("Authorization"):
         hdrs["Authorization"] = request.headers["Authorization"]
+
+    # Trim 0.2s from audio before sending to Whisper (server-side ffmpeg)
+    if subpath == "v1/audio/transcriptions" and request.method == "POST":
+        return proxy_transcription(url, hdrs)
+
     if request.headers.get("Content-Type"):
         hdrs["Content-Type"] = request.headers["Content-Type"]
     r = requests.request(request.method, url, headers=hdrs,
@@ -30,5 +37,48 @@ def proxy(subpath):
     out["Content-Type"] = r.headers.get("Content-Type", "application/json")
     return Response(r.iter_content(8192), status=r.status_code, headers=out)
 
+def proxy_transcription(url, hdrs):
+    audio = request.files.get('file') if request.files else None
+    if not audio:
+        hdrs["Content-Type"] = request.headers.get("Content-Type", "")
+        r = requests.post(url, headers=hdrs, data=request.get_data(), timeout=60)
+        out = dict(CORS)
+        out["Content-Type"] = r.headers.get("Content-Type", "application/json")
+        return Response(r.iter_content(8192), status=r.status_code, headers=out)
+
+    with tempfile.NamedTemporaryFile(suffix='.webm', delete=False) as tmp_in:
+        audio.save(tmp_in)
+        tmp_in_path = tmp_in.name
+
+    tmp_out_path = tmp_in_path + '.wav'
+
+    try:
+        subprocess.run([
+            'ffmpeg', '-y', '-ss', '0.2',
+            '-i', tmp_in_path,
+            '-f', 'wav', tmp_out_path
+        ], check=True, capture_output=True, timeout=15)
+
+        files = {'file': ('audio.wav', open(tmp_out_path, 'rb'), 'audio/wav')}
+        data = {}
+        if request.form.get('model'):
+            data['model'] = request.form['model']
+        if request.form.get('language'):
+            data['language'] = request.form['language']
+
+        r = requests.post(url, headers=hdrs, data=data, files=files, timeout=60)
+    except Exception as e:
+        print(f"[trim] ffmpeg failed: {e}, forwarding original")
+        hdrs["Content-Type"] = request.headers.get("Content-Type", "")
+        r = requests.post(url, headers=hdrs, data=request.get_data(), timeout=60)
+    finally:
+        os.unlink(tmp_in_path)
+        if os.path.exists(tmp_out_path):
+            os.unlink(tmp_out_path)
+
+    out = dict(CORS)
+    out["Content-Type"] = r.headers.get("Content-Type", "application/json")
+    return Response(r.iter_content(8192), status=r.status_code, headers=out)
+
 if __name__ == "__main__":
     app.run(host="127.0.0.1", port=8765)