Update app.py

app.py

@@ -23,7 +23,14 @@ from transformers import (
 from TTS.api import TTS
 
 # Diffusers for sound design generation
-from diffusers import DiffusionPipeline
+from diffusers import DiffusionPipeline, AudioLDMPipeline
+import diffusers
+
+# Monkey-patch: Create a patched pipeline class so that any reference to AudioLDM2Pipeline is resolved correctly.
+class PatchedAudioLDM2Pipeline(AudioLDMPipeline):
+    pass
+
+setattr(diffusers, "AudioLDM2Pipeline", PatchedAudioLDM2Pipeline)
 
 # ---------------------------------------------------------------------
 # Setup Logging and Environment Variables
@@ -100,13 +107,11 @@ def get_tts_model(model_name: str = "tts_models/en/ljspeech/tacotron2-DDC"):
 def get_sound_design_pipeline(model_name: str, token: str):
     """
     Returns a cached DiffusionPipeline for sound design if available;
-    otherwise, it loads and caches the pipeline using the correct pipeline class.
+    otherwise, it loads and caches the pipeline using the patched pipeline class.
     """
     if model_name in SOUND_DESIGN_PIPELINES:
         return SOUND_DESIGN_PIPELINES[model_name]
-    # Import the correct pipeline class from diffusers
-    from diffusers import AudioLDMPipeline
-    pipe = DiffusionPipeline.from_pretrained(model_name, pipeline_class=AudioLDMPipeline, use_auth_token=token)
+    pipe = DiffusionPipeline.from_pretrained(model_name, pipeline_class=PatchedAudioLDM2Pipeline, use_auth_token=token)
     SOUND_DESIGN_PIPELINES[model_name] = pipe
     return pipe
 
@@ -221,7 +226,7 @@ def generate_music(prompt: str, audio_length: int):
 @spaces.GPU(duration=200)
 def generate_sound_design(prompt: str):
     """
-    Generates a sound design audio file based on the provided prompt using Audioldm.
+    Generates a sound design audio file based on the provided prompt using AudioLDM 2.
     Returns the file path to the generated .wav file.
     """
     try: