flac output & max_allowed_segment only for HTDemucs (facebookresearch…

…#500)

Makefile

@@ -16,6 +16,7 @@ test_eval:
 	python3 -m demucs -n demucs_unittest test.mp3
 	python3 -m demucs -n demucs_unittest --two-stems=vocals test.mp3
 	python3 -m demucs -n demucs_unittest --mp3 test.mp3
+	python3 -m demucs -n demucs_unittest --flac --int24 test.mp3
 	python3 -m demucs -n demucs_unittest --int24 --clip-mode clamp test.mp3
 	python3 -m demucs -n demucs_unittest --segment 8 test.mp3
 

demucs/apply.py

@@ -68,7 +68,8 @@ def __init__(self, models: tp.List[Model],
     def max_allowed_segment(self) -> float:
         max_allowed_segment = float('inf')
         for model in self.models:
-            max_allowed_segment = min(max_allowed_segment, float(model.segment))
+            if isinstance(model, HTDemucs):
+                max_allowed_segment = min(max_allowed_segment, float(model.segment))
         return max_allowed_segment
 
     def forward(self, x):

demucs/audio.py

@@ -257,5 +257,7 @@ def save_audio(wav: torch.Tensor,
             encoding = 'PCM_S'
         ta.save(str(path), wav, sample_rate=samplerate,
                 encoding=encoding, bits_per_sample=bits_per_sample)
+    elif suffix == ".flac":
+        ta.save(str(path), wav, sample_rate=samplerate, bits_per_sample=bits_per_sample)
     else:
         raise ValueError(f"Invalid suffix for path: {suffix}")

demucs/separate.py

@@ -104,8 +104,11 @@ def get_parser():
     parser.add_argument("--clip-mode", default="rescale", choices=["rescale", "clamp"],
                         help="Strategy for avoiding clipping: rescaling entire signal "
                              "if necessary  (rescale) or hard clipping (clamp).")
-    parser.add_argument("--mp3", action="store_true",
-                        help="Convert the output wavs to mp3.")
+    format_group = parser.add_mutually_exclusive_group()
+    format_group.add_argument("--flac", action="store_true",
+                              help="Convert the output wavs to flac.")
+    format_group.add_argument("--mp3", action="store_true",
+                              help="Convert the output wavs to mp3.")
     parser.add_argument("--mp3-bitrate",
                         default=320,
                         type=int,
@@ -128,7 +131,7 @@ def main(opts=None):
     except ModelLoadingError as error:
         fatal(error.args[0])
 
-    max_allowed_segment: float = float('inf')
+    max_allowed_segment = float('inf')
     if isinstance(model, HTDemucs):
         max_allowed_segment = float(model.segment)
     elif isinstance(model, BagOfModels):
@@ -170,6 +173,8 @@ def main(opts=None):
 
         if args.mp3:
             ext = "mp3"
+        elif args.flac:
+            ext = "flac"
         else:
             ext = "wav"
         kwargs = {

docs/release.md

@@ -2,11 +2,15 @@
 
 ## V4.0.1a1, TBD
 
+**From this version, Python 3.7 is no longer supported. This is not a problem since the latest PyTorch 2.0.0 no longer support it either.**
+
 Various improvements by @CarlGao4. Support for `segment` param inside of HTDemucs
 model.
 
 Made diffq an optional dependency, with an error message if not installed.
 
+Added output format flac (Free Lossless Audio Codec)
+
 ## V4.0.0, 7th of December 2022
 
 Adding hybrid transformer Demucs model.