update

main.py

@@ -1,6 +1,7 @@
 #!/usr/bin/python3
 # -*- coding: utf-8 -*-
 import argparse
+from functools import lru_cache
 import logging
 from pathlib import Path
 import platform
@@ -71,9 +72,24 @@ denoise_engines = {
             "pretrained_model_path_or_zip_file": (project_path / "trained_models/mpnet-aishell-11-epoch.zip").as_posix()
         }
     },
+    "mpnet-nx-speech-1-epoch": {
+        "infer_cls": InferenceMPNet,
+        "kwargs": {
+            "pretrained_model_path_or_zip_file": (project_path / "trained_models/mpnet-nx-speech-1-epoch.zip").as_posix()
+        }
+    },
 }
 
 
+@lru_cache(maxsize=3)
+def load_denoise_model(infer_engine_param: dict):
+    infer_cls = infer_engine_param["infer_cls"]
+    kwargs = infer_engine_param["kwargs"]
+    infer_engine = infer_cls(**kwargs)
+
+    return infer_engine
+
+
 def when_click_denoise_button(noisy_audio_t, engine: str):
     sample_rate, signal = noisy_audio_t
     logger.info(f"run denoise; engine: {engine}, sample_rate: {sample_rate}, signal dtype: {signal.dtype}, signal shape: {signal.shape}")
@@ -85,9 +101,7 @@ def when_click_denoise_button(noisy_audio_t, engine: str):
         raise gr.Error(f"invalid denoise engine: {engine}.")
 
     try:
-        infer_cls = infer_engine_param["infer_cls"]
-        kwargs = infer_engine_param["kwargs"]
-        infer_engine = infer_cls(**kwargs)
+        infer_engine = load_denoise_model(infer_engine_param)
 
         enhanced_audio = infer_engine.enhancement_by_ndarray(noisy_audio)
         enhanced_audio = np.array(enhanced_audio * (1 << 15), dtype=np.int16)
@@ -117,13 +131,14 @@ def main():
     denoise_engine_choices = list(denoise_engines.keys())
 
     # examples
-    example_zip_file = trained_model_dir / "examples.zip"
-    with zipfile.ZipFile(example_zip_file.as_posix(), "r") as f_zip:
-        out_root = examples_dir
-        if out_root.exists():
-            shutil.rmtree(out_root.as_posix())
-        out_root.mkdir(parents=True, exist_ok=True)
-        f_zip.extractall(path=out_root)
+    if not examples_dir.exists():
+        example_zip_file = trained_model_dir / "examples.zip"
+        with zipfile.ZipFile(example_zip_file.as_posix(), "r") as f_zip:
+            out_root = examples_dir
+            if out_root.exists():
+                shutil.rmtree(out_root.as_posix())
+            out_root.mkdir(parents=True, exist_ok=True)
+            f_zip.extractall(path=out_root)
 
     # examples
     examples = list()

toolbox/torchaudio/models/clean_unet/modeling_clean_unet.py

@@ -7,6 +7,10 @@ https://github.com/nvidia/cleanunet
 
 https://huggingface.co/spaces/fsoft-ai-center/Speech-Enhancement/blob/main/src/model.py
 
+支持流式改造。
+
+https://github.com/francislr/clean-unet-inference
+
 """
 import os
 from typing import Optional, Union

toolbox/torchaudio/models/conv_tasnet/__init__.py

@@ -0,0 +1,6 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+
+
+if __name__ == '__main__':
+    pass

toolbox/torchaudio/models/conv_tasnet/modeling_conv_tasnet.py

@@ -0,0 +1,9 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+"""
+https://github.com/kaituoxu/Conv-TasNet/blob/master/src/conv_tasnet.py
+"""
+
+
+if __name__ == '__main__':
+    pass