Fix STT health check incorrectly marking Open WebUI as whisper.cpp

Port 8080 is Open WebUI — it passes /health + /v1/models checks but
returns 405 on POST /v1/audio/transcriptions. Updated probe logic to:
- treat 405 as 'endpoint missing, try next path'
- treat non-JSON 500 as broken, JSON-500 with detail as 'audio too short' (ok)
- use 500ms silence WAV instead of 1-frame (too tiny for alignment models)

Changed whisper.cpp default from :8080 to :8085 to avoid clash with
Open WebUI. Updated s-llms.html placeholder and code snippet accordingly.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

server.py

@@ -48,7 +48,7 @@ _NVIDIA_CLONE_DEFAULT = os.environ.get("NVIDIA_TTS_CLONE_URL", "http://host.dock
 _NVIDIA_ZEROSHOT_DEFAULT = os.environ.get("NVIDIA_ZEROSHOT_TTS_URL", _NVIDIA_CLONE_DEFAULT)
 _NVIDIA_FLOW_DEFAULT = os.environ.get("NVIDIA_FLOW_TTS_URL", "http://host.docker.internal:8094")
 _FASTER_WHISPER_DEFAULT = os.environ.get("FASTER_WHISPER_URL", "http://host.docker.internal:8000")
-_WHISPER_CPP_DEFAULT    = os.environ.get("WHISPER_CPP_URL",    "http://host.docker.internal:8080")
+_WHISPER_CPP_DEFAULT    = os.environ.get("WHISPER_CPP_URL",    "http://host.docker.internal:8085")
 _GROQ_STT_ENDPOINT      = "https://api.groq.com/openai/v1"
 _KOKORO_DEFAULT         = os.environ.get("KOKORO_URL",         "http://host.docker.internal:8880/v1")
 _VIBEVOICE_DEFAULT      = os.environ.get("VIBEVOICE_URL",      "http://192.168.178.8:8027")
@@ -1903,15 +1903,15 @@ def _stt_backend_api_key(settings: dict, backend: str) -> str:
     return settings.get("whisper_api_key", "").strip()
 
 
-def _make_minimal_wav() -> bytes:
+def _make_minimal_wav(duration_ms: int = 500) -> bytes:
-    """44-byte WAV header with 1 frame of silence — smallest valid WAV."""
+    """Minimal WAV: mono 16-bit 16kHz silence of given duration."""
-    import struct
+    sample_rate = 16000
-    num_frames = 1
+    num_frames = sample_rate * duration_ms // 1000
-    data = b"\x00\x00"
+    data = b"\x00\x00" * num_frames
     header = struct.pack(
         "<4sI4s4sIHHIIHH4sI",
         b"RIFF", 36 + len(data), b"WAVE",
-        b"fmt ", 16, 1, 1, 16000, 32000, 2, 16,
+        b"fmt ", 16, 1, 1, sample_rate, sample_rate * 2, 2, 16,
         b"data", len(data),
     )
     return header + data
@@ -1942,9 +1942,11 @@ def _stt_backend_health(url: str) -> tuple[bool, list[str]]:
     if not ok:
         return False, models
     # Verify the transcription endpoint actually works (catches broken builds like
-    # CTranslate2 containers compiled without CUDA that pass /health but fail on load)
+    # CTranslate2 containers compiled without CUDA that pass /health but fail on load,
+    # and LLM servers that expose /v1/models but no transcription endpoint)
     try:
         wav = _make_minimal_wav()
+        endpoint_found = False
         for path in ("/v1/audio/transcriptions", "/transcribe"):
             try:
                 r = requests.post(
@@ -1953,14 +1955,29 @@ def _stt_backend_health(url: str) -> tuple[bool, list[str]]:
                     data={"model": "whisper-1", "response_format": "text"},
                     timeout=8,
                 )
+                if r.status_code in {404, 405}:
+                    continue  # try next path — endpoint missing or wrong method
                 if r.status_code == 500:
-                    ok = False  # backend is broken (model load failed etc.)
+                    # Distinguish backend crash from "couldn't process empty audio":
-                break  # 200, 400, 404, 422 all mean the endpoint exists and model loaded
+                    # a JSON detail that isn't the generic FastAPI text means the
+                    # model loaded and responded — treat as ok (audio was just bad)
+                    try:
+                        detail = r.json().get("detail", "")
+                        ok = bool(detail) and detail.lower() != "internal server error"
+                    except Exception:
+                        ok = False  # non-JSON 500 → backend likely broken
+                else:
+                    ok = True   # 200, 400, 415, 422 → endpoint works
+                endpoint_found = True
+                break
             except requests.exceptions.ConnectionError:
                 ok = False
+                endpoint_found = True
                 break
             except Exception:
-                pass  # timeout or other — don't mark as broken, just skip
+                pass  # timeout — don't penalise slow model loads
+        if not endpoint_found:
+            ok = False  # all paths returned 404/405 — not an STT backend
     except Exception:
         pass
     return ok, models

static/sections/s-llms.html

@@ -350,7 +350,7 @@ Check "Enable CORS" for browser access</pre>
             <p class="llm-local-desc">Minimal C++ Whisper with a built-in HTTP server. CPU or Metal/CUDA. Low memory, fast on consumer hardware.</p>
             <div class="llm-local-url">
               <span class="llm-local-url-label">URL</span>
-              <input class="llm-local-url-inp" type="text" placeholder="http://localhost:8080" data-llm-local-key="whisper-cpp" data-llm-local-default="http://localhost:8080" spellcheck="false">
+              <input class="llm-local-url-inp" type="text" placeholder="http://localhost:8085" data-llm-local-key="whisper-cpp" data-llm-local-default="http://localhost:8085" spellcheck="false">
               <button class="llm-local-ping" data-ping-key="whisper-cpp" title="Test connection">Connect</button>
             </div>
             <div class="llm-local-actions">
@@ -359,12 +359,12 @@ Check "Enable CORS" for browser access</pre>
             <div class="llm-local-snippet">
               <div class="llm-snippet-bar">
                 <span>Build &amp; run</span>
-                <button class="llm-copy-btn" data-copy="git clone https://github.com/ggml-org/whisper.cpp&#10;cd whisper.cpp && cmake -B build && cmake --build build -j&#10;./build/bin/whisper-server -m models/ggml-large-v3.bin --port 8080">Copy</button>
+                <button class="llm-copy-btn" data-copy="git clone https://github.com/ggml-org/whisper.cpp&#10;cd whisper.cpp && cmake -B build && cmake --build build -j&#10;./build/bin/whisper-server -m models/ggml-large-v3.bin --port 8085">Copy</button>
               </div>
               <pre>git clone https://github.com/ggml-org/whisper.cpp
 cd whisper.cpp && cmake -B build && cmake --build build -j
 ./build/bin/whisper-server \
-  -m models/ggml-large-v3.bin --port 8080</pre>
+  -m models/ggml-large-v3.bin --port 8085</pre>
             </div>
             <a class="llm-local-link" href="https://github.com/ggml-org/whisper.cpp" target="_blank" rel="noopener">github.com/ggml-org/whisper.cpp <span class="mdi mdi-open-in-new link-icon"></span></a>
           </div>