feat(tts): mixed-language synthesis via <en> tags

Parse <en>word</en> markers in text, synthesise English segments with
en-US-GuyNeural and German segments with de-DE-ConradNeural, then
ffmpeg-concat into a single MP3. Fallback to plain synthesis if no tags.

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

compliance-tts-service/main.py

@@ -141,11 +141,16 @@ EDGE_TTS_VOICES = {
     "en": "en-US-GuyNeural",
 }
 
+# Matches <en>word</en> markers used to force English pronunciation in German text
+_EN_TAG_RE = re.compile(r'<en>(.*?)</en>', re.DOTALL)
+
 
 async def _edge_tts_synthesize(text: str, language: str, output_path: str) -> bool:
-    """Synthesize using Edge TTS (Microsoft Neural Voices). Returns True on success."""
+    """Synthesize using Edge TTS. Handles <en>…</en> mixed-language markers."""
     try:
         import edge_tts
+        if '<en>' in text:
+            return await _edge_tts_mixed(text, language, output_path)
         voice = EDGE_TTS_VOICES.get(language, EDGE_TTS_VOICES["de"])
         communicate = edge_tts.Communicate(text, voice)
         await communicate.save(output_path)
@@ -155,6 +160,45 @@ async def _edge_tts_synthesize(text: str, language: str, output_path: str) -> bo
         return False
 
 
+async def _edge_tts_mixed(text: str, base_lang: str, output_path: str) -> bool:
+    """Split on <en> tags, synthesise each segment with the right voice, concat."""
+    import edge_tts, shutil, tempfile as tf
+    segments: list[tuple[str, str]] = []
+    last = 0
+    for m in _EN_TAG_RE.finditer(text):
+        if m.start() > last:
+            segments.append((base_lang, text[last:m.start()].strip()))
+        segments.append(("en", m.group(1).strip()))
+        last = m.end()
+    if last < len(text):
+        segments.append((base_lang, text[last:].strip()))
+    segments = [(lang, t) for lang, t in segments if t]
+
+    tmpdir = tf.mkdtemp()
+    try:
+        seg_files = []
+        for i, (lang, seg_text) in enumerate(segments):
+            voice = EDGE_TTS_VOICES.get(lang, EDGE_TTS_VOICES["de"])
+            seg_path = os.path.join(tmpdir, f"seg_{i:04d}.mp3")
+            await edge_tts.Communicate(seg_text, voice).save(seg_path)
+            seg_files.append(seg_path)
+
+        list_file = os.path.join(tmpdir, "list.txt")
+        with open(list_file, "w") as f:
+            for sf in seg_files:
+                f.write(f"file '{sf}'\n")
+        proc = subprocess.run(
+            ["ffmpeg", "-y", "-f", "concat", "-safe", "0", "-i", list_file, "-c", "copy", output_path],
+            capture_output=True, text=True, timeout=60,
+        )
+        return proc.returncode == 0
+    except Exception as e:
+        logger.warning(f"Mixed TTS failed: {e}")
+        return False
+    finally:
+        shutil.rmtree(tmpdir, ignore_errors=True)
+
+
 @app.post("/synthesize-direct")
 async def synthesize_direct(req: SynthesizeDirectRequest):
     """Synthesize text and return MP3 audio directly (no MinIO upload).