feat: add speech noise separation

experiments/bs-roformer-vocal-separation/app.py

@@ -26,6 +26,7 @@ MODES = {
     "vocals": {"model": MODEL, "stems": ("vocals", "instrumental"), "archive": "athena-vocals-instrumental.zip", "engine": "mdxc"},
     "four_stem": {"model": "htdemucs_ft.yaml", "stems": ("vocals", "drums", "bass", "other"), "archive": "athena-4-stems.zip", "engine": "demucs"},
     "six_stem": {"model": "htdemucs_6s.yaml", "stems": ("vocals", "drums", "bass", "guitar", "piano", "other"), "archive": "athena-6-stems-experimental.zip", "engine": "demucs"},
+    "speech": {"model": "MossFormer2_SE_48K", "stems": ("speech", "noise"), "archive": "athena-sprache-und-hintergrund.zip", "engine": "clearvoice"},
 }
 TARGETS = {
     "vocals": {"mode": "vocals", "stem": "vocals", "remainder": "instrumental", "archive": "athena-gesang-und-rest.zip", "rest_file": "instrumental.flac"},
@@ -34,6 +35,7 @@ TARGETS = {
     "guitar": {"mode": "six_stem", "stem": "guitar", "archive": "athena-gitarre-und-rest.zip", "rest_file": "rest-ohne-gitarre.flac"},
     "piano": {"mode": "six_stem", "stem": "piano", "archive": "athena-piano-und-rest.zip", "rest_file": "rest-ohne-piano.flac"},
     "other": {"mode": "six_stem", "stem": "other", "archive": "athena-sonstiges-und-rest.zip", "rest_file": "rest-ohne-sonstiges.flac"},
+    "speech": {"mode": "speech", "stem": "speech", "remainder": "noise", "archive": "athena-sprache-und-hintergrund.zip", "rest_file": "hintergrund-ohne-sprache.flac"},
 }
 
 app = FastAPI(title="Athena Stem Separator", version="2.0")
@@ -46,7 +48,14 @@ def index() -> str:
 
 @app.get("/health")
 def health() -> dict:
-    available = {name: (MODEL_DIR / mode["model"]).exists() for name, mode in MODES.items()}
+    available = {
+        name: (
+            (MODEL_DIR / "clearvoice" / mode["model"] / "last_best_checkpoint").exists()
+            if mode["engine"] == "clearvoice"
+            else (MODEL_DIR / mode["model"]).exists()
+        )
+        for name, mode in MODES.items()
+    }
     return {
         "status": "ok" if all(available.values()) else "starting",
         "models": {name: mode["model"] for name, mode in MODES.items()},
@@ -62,6 +71,20 @@ def _cleanup(path: Path) -> None:
 
 
 def _run_separator(input_path: Path, output_dir: Path, mode: dict) -> None:
+    if mode["engine"] == "clearvoice":
+        completed = subprocess.run(
+            [
+                "/opt/clearvoice-venv/bin/python", "/app/speech_enhance.py", str(input_path),
+                str(output_dir / "speech.flac"), str(output_dir / "noise.flac"),
+            ],
+            capture_output=True,
+            text=True,
+            timeout=7200,
+        )
+        if completed.returncode:
+            detail = (completed.stderr or completed.stdout or "unknown ClearVoice error")[-4000:]
+            raise RuntimeError(detail)
+        return
     args = [
         "audio-separator", str(input_path),
         "--model_filename", mode["model"],