From 6ce8d34317f85c602f36c153f00460a23f039e7e Mon Sep 17 00:00:00 2001
From: RubenRWU <ruben.fischer@rwu.de>
Date: Tue, 7 Apr 2026 15:28:56 +0200
Subject: [PATCH] tried to fix transcibe failures

---
 .env.example |  1 +
 main.py      | 41 +++++++++++++++++++++++++++++++----------
 2 files changed, 32 insertions(+), 10 deletions(-)

diff --git a/.env.example b/.env.example
index 4eb1446..737a6d6 100644
--- a/.env.example
+++ b/.env.example
@@ -5,6 +5,7 @@ SMTP_PASSWORD=your_smtp_password_here
 SMTP_FROM=Meeting Transcription Service
 AZURE_SPEECH_ENDPOINT=https://your-speech-resource.cognitiveservices.azure.com/
 AZURE_SPEECH_KEY=your_azure_speech_key_here
+AZURE_SPEECH_LOCALE=de-DE
 AZURE_SPEECH_DIARIZATION_ENABLED=true
 AZURE_SPEECH_MAX_SPEAKERS=8
 AZURE_SPEECH_LLM_PROMPT=Transkribiere dieses Geschäftsmeeting mit hoher Genauigkeit auf Deutsch. Bewahre Formulierung und Bedeutung so genau wie möglich. Fasse nichts zusammen, schreibe nichts um und ergänze keine Erklärungen. Setze saubere Interpunktion und klare Satzgrenzen. Trenne Sprecherwechsel gemäß der Audioaufnahme. Bewahre Fachbegriffe, Eigennamen, Akronyme, Zahlen, Währungen, Datumsangaben, Aufgaben und fremdsprachige Begriffe exakt. Gib nur das Transkript aus.
diff --git a/main.py b/main.py
index c36e6e5..555fe05 100644
--- a/main.py
+++ b/main.py
@@ -108,15 +108,8 @@ def is_azure_direct_upload_supported(file_path: Path, fmt: str) -> bool:
         ".mp3",
         ".ogg",
         ".flac",
-        ".webm",
-        ".wma",
-        ".aac",
-        ".amr",
-        ".speex",
-        ".m4a",
-        ".mp4",
     }
-    supported_formats = {"wav", "mp3", "ogg", "flac", "webm"}
+    supported_formats = {"wav", "mp3", "ogg", "flac"}
     return file_path.suffix.lower() in supported_exts or fmt in supported_formats
 
 
@@ -150,6 +143,15 @@ def _build_azure_definition() -> dict:
         },
         "profanityFilterMode": os.getenv("AZURE_SPEECH_PROFANITY_MODE", "Masked"),
     }
+    locales_raw = (
+        os.getenv("AZURE_SPEECH_LOCALES", "").strip()
+        or os.getenv("AZURE_SPEECH_LOCALE", "").strip()
+    )
+    if locales_raw:
+        locales = [item.strip() for item in locales_raw.split(",") if item.strip()]
+        if locales:
+            definition["locales"] = locales
+
     prompt = os.getenv("AZURE_SPEECH_LLM_PROMPT", "").strip()
     if prompt:
         definition["enhancedMode"]["prompt"] = [prompt]
@@ -175,7 +177,17 @@ def _encode_multipart_form(fields: dict[str, str], file_field: str, file_path: P
         body.extend(value.encode("utf-8"))
         body.extend(b"\r\n")
 
-    mime_type = "audio/wav" if file_path.suffix.lower() == ".wav" else "application/octet-stream"
+    mime_types = {
+        ".wav": "audio/wav",
+        ".mp3": "audio/mpeg",
+        ".ogg": "audio/ogg",
+        ".flac": "audio/flac",
+        ".webm": "audio/webm",
+        ".aac": "audio/aac",
+        ".m4a": "audio/mp4",
+        ".mp4": "audio/mp4",
+    }
+    mime_type = mime_types.get(file_path.suffix.lower(), "application/octet-stream")
     body.extend(f"--{boundary}\r\n".encode("utf-8"))
     body.extend(
         (
@@ -244,7 +256,16 @@ def _format_azure_transcript(result: dict) -> str:
     if merged_text:
         return f"SPRECHER: {merged_text}"
 
-    raise RuntimeError("Azure Speech lieferte kein Transkript zurück")
+    diagnostics = {
+        "durationMilliseconds": result.get("durationMilliseconds"),
+        "keys": sorted(result.keys()),
+    }
+    preview = json.dumps(result, ensure_ascii=False)[:1200]
+    raise RuntimeError(
+        "Azure Speech lieferte kein Transkript zurück. "
+        f"Diagnose={json.dumps(diagnostics, ensure_ascii=False)} "
+        f"Antwort={preview}"
+    )
 
 
 def transcribe_audio(file_path: Path) -> str: