feat: overlap-window speechbrain enhance, hop 200ms / window 1s

run_worker.py

@@ -35,7 +35,8 @@ def main() -> int:
     ap.add_argument("--enhance-mode", default="auto")
     ap.add_argument("--enhance-model",
                     default="speechbrain/metricgan-plus-voicebank")
-    ap.add_argument("--enhance-chunk-s", type=float, default=0.2)
+    ap.add_argument("--enhance-chunk-s", type=float, default=1.0)
+    ap.add_argument("--enhance-hop-s", type=float, default=0.2)
     args = ap.parse_args()
 
     cam = json.loads(args.camera_json)
@@ -57,6 +58,8 @@ def main() -> int:
             mode=args.enhance_mode,
             model_source=args.enhance_model,
             sample_rate=args.audio_rate,
+            chunk_s=args.enhance_chunk_s,
+            hop_s=args.enhance_hop_s,
         )
         log.info("audio backend: %s", cleaner.backend)
 
@@ -125,8 +128,8 @@ def main() -> int:
             atrack = rtc.LocalAudioTrack.create_audio_track(
                 f"{identity}-audio", audio_source)
 
-            chunk_s = args.enhance_chunk_s
-            chunk = int(chunk_s * args.audio_rate)  # samples per channel
+            hop_s = args.enhance_hop_s
+            chunk = int(hop_s * args.audio_rate)  # samples per hop
 
             async def audio_loop():
                 assert aproc is not None and atrack is not None