feat: replace torchaudio with torchcodec for audio decoding

app.py

@@ -4,7 +4,7 @@ from importlib.resources import files
 import gradio as gr
 import soundfile as sf
 import torch
-import torchaudio
+import torchcodec
 from cached_path import cached_path
 from omegaconf import OmegaConf
 
@@ -164,7 +164,7 @@ def infer(
         with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
             sf.write(f.name, final_wave, final_sample_rate)
             remove_silence_for_generated_wav(f.name)
-            final_wave, _ = torchaudio.load(f.name)
+            final_wave = torchcodec.decoders.AudioDecoder(f.name).get_all_samples().data
         final_wave = final_wave.squeeze().cpu().numpy()
 
     # Save the spectrogram

requirements.txt

@@ -1,3 +1,4 @@
 omegaconf
 opencc
+torchcodec
 git+https://github.com/SWivid/F5-TTS.git