Bug fix for repitching & tempo automix errors

README.md

@@ -138,7 +138,7 @@ If you have anaconda installed, you can run from the root of this repository:
 conda env update -f environment-cpu.yml  # if you don't have GPUs
 conda env update -f environment-cuda.yml # if you have GPUs
 conda activate demucs
-pip install -e .
+pip install -e .[dev]
 ```
 
 This will create a `demucs` environment with all the dependencies installed.

requirements.txt

@@ -17,3 +17,4 @@ torchaudio>=0.8,<2.1
 tqdm
 treetable
 soundfile>=0.10.3;sys_platform=="win32"
+librosa @ git+https://github.com/librosa/librosa.git

tools/automix.py

@@ -205,12 +205,18 @@ def find_candidate(spec_ref, catalog, pitch_match=True):
 def get_part(spec, source, dt, dp):
     """Apply given delta of tempo and delta of pitch to a stem."""
     wav = spec.track[source]
-    if dt or dp:
-        wav = repitch(wav, dp, dt * 100, samplerate=SR, voice=source == 3)
+    if dt != 0 or dp != 0:  # Check if there's any change to apply
+        # Ensure 'dt' is a scalar if it's an array
+        if isinstance(dt, np.ndarray) and dt.size == 1:
+            dt = float(dt.item())  # Convert numpy array to Python scalar
+        # Convert tempo change from relative change (e.g., -0.12 for 88%) to percentage change expected by `repitch`
+        tempo_percentage_change = dt * 100  # Convert to percentage
+        # Apply pitch and tempo changes
+        wav = repitch(wav, dp, tempo_percentage_change, voice=source == 3, samplerate=SR)
+        # Adjust onsets according to new tempo
         spec = spec._replace(onsets=spec.onsets / (1 + dt))
     return wav, spec
 
-
 def build_track(ref_index, catalog):
     """Given the reference track index and a catalog of track, builds
     a completely new track. One of the source at random from the ref track will