Finalize 0.0.1

Author: enhuiz

config/enhancer_stage1.yaml

@@ -1,5 +1,4 @@
 lcfm_training_mode: ae
 load_fg_only: true
 batch_size_per_gpu: 16
-# Uncomment this if you are training your model
-# denoiser_run_dir: runs/denoiser
+denoiser_run_dir: runs/denoiser

config/enhancer_stage2.yaml

@@ -4,6 +4,5 @@ training_seconds: 3.0
 gan_training_start_step: null
 lcfm_z_scale: 6
 praat_augment_prob: 0.2
-# Uncomment this if you are training your model
-# enhancer_stage1_run_dir: runs/enhancer_stage1
-# denoiser_run_dir: runs/denoiser
+denoiser_run_dir: runs/denoiser
+enhancer_stage1_run_dir: runs/enhancer_stage1

resemble_enhance/data/distorter/custom.py

@@ -42,6 +42,8 @@ def _sample_rir(self):
         return rir
 
     def apply(self, wav, sr):
+        # ref: https://github.com/haoheliu/voicefixer_main/blob/b06e07c945ac1d309b8a57ddcd599ca376b98cd9/dataloaders/augmentation/magical_effects.py#L158
+
         if len(self.rir_paths) == 0:
             return wav
 

resemble_enhance/enhancer/__main__.py

@@ -1,5 +1,6 @@
 import argparse
 import random
+import time
 from pathlib import Path
 
 import torch
@@ -70,6 +71,8 @@ def main():
 
     args = parser.parse_args()
 
+    start_time = time.perf_counter()
+
     run_dir = args.run_dir
 
     paths = sorted(args.in_dir.glob(f"**/*{args.suffix}"))
@@ -111,6 +114,10 @@ def main():
         out_path.parent.mkdir(parents=True, exist_ok=True)
         torchaudio.save(out_path, hwav[None], sr)
 
+    # Cool emoji effect saying the job is done
+    elapsed_time = time.perf_counter() - start_time
+    print(f"🌟 Enhancement done! {len(paths)} files processed in {elapsed_time:.2f}s")
+
 
 if __name__ == "__main__":
     main()

setup.py

@@ -10,8 +10,8 @@ def shell(*args):
     return out.decode("ascii").strip()
 
 
-def write_version(version_core, dev=True):
-    if dev:
+def write_version(version_core, pre_release=True):
+    if pre_release:
         last_commit_time = shell("git", "log", "-1", "--format=%cd", "--date=iso-strict")
         last_commit_time = datetime.strptime(last_commit_time, "%Y-%m-%dT%H:%M:%S%z")
         last_commit_time = last_commit_time.astimezone(timezone.utc)
@@ -36,7 +36,7 @@ def write_version(version_core, dev=True):
 setup(
     name="resemble-enhance",
     python_requires=">=3.10",
-    version=write_version("0.0.1"),
+    version=write_version("0.0.1", pre_release=False),
     description="Speech denoising and enhancement with deep learning",
     long_description=long_description,
     long_description_content_type="text/markdown",