[Admin maintenance] Support new ZeroGPU hardware

app.py

@@ -15,7 +15,7 @@ from typing import Union
 from diffusers.utils.torch_utils import randn_tensor
 from tqdm import tqdm
 from TangoFlux import TangoFluxInference
-import torchaudio
+import soundfile as sf
 
 # Define the description text
 description_text = """
@@ -43,7 +43,15 @@ def gradio_generate(prompt, steps, guidance, duration):
     output = tangoflux.generate(prompt, steps=steps, guidance_scale=guidance, duration=duration)
     filename = 'temp.wav'
     output = output[:,:int(duration*44100)]
-    torchaudio.save(filename, output, 44100)
+    # torchaudio.save replaced with soundfile.write (Step 5: torchaudio 2.10+ delegates .save to torchcodec)
+    # output is (C, F) torch tensor; soundfile expects (F,) for mono or (F, C) for multi-channel
+    audio = output.detach().cpu().numpy()
+    if audio.ndim == 2:
+        if audio.shape[0] == 1:
+            audio = audio[0]
+        else:
+            audio = audio.T  # (C, F) -> (F, C)
+    sf.write(filename, audio, 44100)
     return filename
 
 # Create custom interface with HTML badges

requirements.txt

@@ -1,7 +1,4 @@
-torch==2.4.0
-torchaudio===2.4.0
 torchlibrosa==0.1.0
-torchvision==0.19.0
 transformers==4.44.0
 diffusers==0.32.0
 accelerate==0.34.2
@@ -9,3 +6,5 @@ datasets==2.21.0
 librosa
 tqdm
 wavio==0.0.7
+soundfile
+pydub