Add private realtime Talk for OpenClaw

router/stt_worker.py

@@ -3,7 +3,8 @@
 STT-Worker – langlebiger Whisper-Transkriptions-Service (CPU-only).
 
 Liest Audio-Dateien (WAV, MP3, OGG, FLAC, WebM/Opus via ffmpeg),
-transkribiert sie mit whisper.cpp (whisper-cli) und liefert JSON-Text.
+transkribiert sie mit einem dauerhaft geladenen whisper.cpp-Server (mit
+whisper-cli als Rückfallweg) und liefert JSON-Text.
 
 Konfiguration über Umgebungsvariablen:
   WHISPER_HOST       Bind-Adresse (Default: 127.0.0.1)
@@ -28,6 +29,7 @@ import sys
 import tempfile
 import time
 import uuid
+import urllib.request
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 
 # ---------------------------------------------------------------------------
@@ -45,6 +47,7 @@ WHISPER_MODEL = os.environ.get(
 )
 WHISPER_THREADS = int(os.environ.get("WHISPER_THREADS", "8"))
 WHISPER_LANGUAGE = os.environ.get("WHISPER_LANGUAGE", "de")
+WHISPER_SERVER_URL = os.environ.get("WHISPER_SERVER_URL", "").rstrip("/")
 FFMPEG_BIN = os.environ.get("FFMPEG_BIN", "/usr/bin/ffmpeg")
 LOG_LEVEL = os.environ.get("LOG_LEVEL", "INFO")
 
@@ -139,13 +142,22 @@ def transcribe(
     temperature: float | None = None,
 ) -> dict:
     """
-    Führt die Transkription mit whisper-cli aus.
+    Führt die Transkription bevorzugt über den persistenten whisper.cpp-
+    Server aus. Dadurch wird das Modell nicht pro Aufnahme neu geladen.
     Liefert dict mit 'text' und Metadaten.
     """
     lang = language or WHISPER_LANGUAGE
     if lang == "auto":
         lang = "auto"
 
+    if WHISPER_SERVER_URL:
+        return _transcribe_via_server(
+            audio_path,
+            language=lang,
+            prompt=prompt,
+            temperature=temperature,
+        )
+
     out_prefix = f"/tmp/stt_{uuid.uuid4().hex[:12]}"
     out_json = out_prefix + ".json"
 
@@ -217,6 +229,65 @@ def transcribe(
     return result
 
 
+def _transcribe_via_server(
+    audio_path: str,
+    language: str,
+    prompt: str | None = None,
+    temperature: float | None = None,
+) -> dict:
+    """Sendet eine Aufnahme an den bereits geladenen whisper-server."""
+    boundary = f"----athena-whisper-{uuid.uuid4().hex}"
+    chunks: list[bytes] = []
+
+    def add_field(name: str, value: str) -> None:
+        chunks.extend([
+            f"--{boundary}\r\n".encode(),
+            f'Content-Disposition: form-data; name="{name}"\r\n\r\n'.encode(),
+            value.encode("utf-8"),
+            b"\r\n",
+        ])
+
+    with open(audio_path, "rb") as f:
+        audio = f.read()
+    chunks.extend([
+        f"--{boundary}\r\n".encode(),
+        b'Content-Disposition: form-data; name="file"; filename="audio.wav"\r\n',
+        b"Content-Type: audio/wav\r\n\r\n",
+        audio,
+        b"\r\n",
+    ])
+    add_field("response_format", "json")
+    add_field("language", language)
+    if prompt:
+        add_field("prompt", prompt)
+    if temperature is not None:
+        add_field("temperature", str(temperature))
+    chunks.append(f"--{boundary}--\r\n".encode())
+
+    request = urllib.request.Request(
+        f"{WHISPER_SERVER_URL}/inference",
+        data=b"".join(chunks),
+        headers={"Content-Type": f"multipart/form-data; boundary={boundary}"},
+        method="POST",
+    )
+    t0 = time.monotonic()
+    with urllib.request.urlopen(request, timeout=300) as response:
+        payload = json.loads(response.read().decode("utf-8"))
+    elapsed = time.monotonic() - t0
+    text = payload.get("text", "") if isinstance(payload, dict) else ""
+    result = {
+        "text": text.strip(),
+        "language": language,
+        "duration_ms": int(elapsed * 1000),
+        "engine": "whisper-server",
+    }
+    log.info(
+        "Transkription (persistent): %d ms, %d Zeichen, Sprache=%s",
+        result["duration_ms"], len(result["text"]), language,
+    )
+    return result
+
+
 # ---------------------------------------------------------------------------
 # HTTP-Handler
 # ---------------------------------------------------------------------------
@@ -300,6 +371,8 @@ class STTHandler(BaseHTTPRequestHandler):
                 "whisper_cli_exists": cli_ok,
                 "threads": WHISPER_THREADS,
                 "language": WHISPER_LANGUAGE,
+                "server_url": WHISPER_SERVER_URL or None,
+                "persistent_model": bool(WHISPER_SERVER_URL),
                 "ffmpeg": FFMPEG_BIN,
                 "ffmpeg_exists": os.path.isfile(FFMPEG_BIN),
             })