Improved --single_stem parameter to be more intuitive, expecting "instrumental" or "vocals" rather than "primary" etc.

Author: beveradb

audio_separator/separator.py

@@ -86,7 +86,9 @@ def __init__(
 
         self.output_single_stem = output_single_stem
         if output_single_stem is not None:
-            self.logger.debug(f"Single stem output requested, only one output file will be written: {output_single_stem}")
+            if output_single_stem.lower() not in {"instrumental", "vocals"}:
+                raise Exception("output_single_stem must be either 'instrumental' or 'vocals'")
+            self.logger.debug(f"Single stem output requested, only one output file ({output_single_stem}) will be written")
 
         self.chunks = 0
         self.margin = 44100
@@ -170,15 +172,15 @@ def separate(self):
 
         output_files = []
 
-        if self.output_single_stem != "secondary":
+        if not self.output_single_stem or self.output_single_stem.lower() == self.primary_stem.lower():
             self.logger.info(f"Saving {self.primary_stem} stem...")
             primary_stem_path = os.path.join(f"{self.audio_file_base}_({self.primary_stem})_{self.model_name}.{self.output_format.lower()}")
             if not isinstance(self.primary_source, np.ndarray):
                 self.primary_source = spec_utils.normalize(self.logger, source, self.normalization_enabled).T
             self.write_audio(primary_stem_path, self.primary_source, samplerate)
             output_files.append(primary_stem_path)
 
-        if self.output_single_stem != "primary":
+        if not self.output_single_stem or self.output_single_stem.lower() == self.secondary_stem.lower():
             self.logger.info(f"Saving {self.secondary_stem} stem...")
             secondary_stem_path = os.path.join(
                 f"{self.audio_file_base}_({self.secondary_stem})_{self.model_name}.{self.output_format.lower()}"

audio_separator/utils/cli.py

@@ -75,7 +75,7 @@ def main():
     parser.add_argument(
         "--single_stem",
         default=None,
-        help="Optional: output only single stem, primary or secondary. With the default model, primary is the instrumental. Example: --single_stem=primary",
+        help="Optional: output only single stem, either instrumental or vocals. Example: --single_stem=instrumental",
     )
 
     args = parser.parse_args()

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "audio-separator"
-version = "0.6.2"
+version = "0.6.3"
 description = "Easy to use vocal separation on CLI or as a python package, using the amazing MDX-Net models from UVR trained by @Anjok07"
 authors = ["Andrew Beveridge <[email protected]>"]
 license = "MIT"