feat: OpenVINO CPU MetricGAN mask; 100 ms hop

Per-worker OpenVINO CPU runs the MetricGAN BLSTM mask; SpeechBrain
keeps STFT/ISTFT. ENHANCE_INFER=auto with torch fallback. Hop is
0.1 s after isolated last_ms ~6 ms (model+ov+beamform).

config.py

@@ -41,7 +41,8 @@ class Config:
     enhance_mode: str = "auto"  # auto | force | off
     enhance_model: str = "speechbrain/metricgan-plus-voicebank"
     enhance_chunk_s: float = 1.0
-    enhance_hop_s: float = 0.2
+    enhance_hop_s: float = 0.1
+    enhance_infer: str = "auto"  # auto | openvino | torch
     # SpeechBrain Delay-and-Sum (GCC-PHAT) on C920 stereo before enhance
     beamform_mode: str = "auto"  # auto | force | off
     torch_num_threads: int = 1
@@ -118,7 +119,8 @@ def load_config(env_path: Optional[Path] = None) -> Config:
             "ENHANCE_MODEL", "speechbrain/metricgan-plus-voicebank"
         ),
         enhance_chunk_s=float(os.environ.get("ENHANCE_CHUNK_S", "1.0")),
-        enhance_hop_s=float(os.environ.get("ENHANCE_HOP_S", "0.2")),
+        enhance_hop_s=float(os.environ.get("ENHANCE_HOP_S", "0.1")),
+        enhance_infer=os.environ.get("ENHANCE_INFER", "auto").lower(),
         beamform_mode=os.environ.get("BEAMFORM_MODE", "auto").lower(),
         torch_num_threads=int(os.environ.get("TORCH_NUM_THREADS", "1")),
         beamform_tdoa_every=int(os.environ.get("BEAMFORM_TDOA_EVERY", "5")),
@@ -146,6 +148,9 @@ def load_config(env_path: Optional[Path] = None) -> Config:
 
     if cfg.enhance_mode not in ("auto", "force", "off"):
         raise ValueError(f"ENHANCE_MODE must be auto|force|off, got {cfg.enhance_mode}")
+    if cfg.enhance_infer not in ("auto", "openvino", "torch"):
+        raise ValueError(
+            f"ENHANCE_INFER must be auto|openvino|torch, got {cfg.enhance_infer}")
     if cfg.beamform_mode not in ("auto", "force", "off"):
         raise ValueError(
             f"BEAMFORM_MODE must be auto|force|off, got {cfg.beamform_mode}")