update apps to incude latent diffusion

app.py

@@ -15,28 +15,30 @@ def generate_spectrogram_audio_and_loop(model_id):
     return image, (sample_rate, audio), (sample_rate, loop)
 
 
-demo = gr.Interface(fn=generate_spectrogram_audio_and_loop,
-                    title="Audio Diffusion",
-                    description="Generate audio using Huggingface diffusers.\
-        This takes about 20 minutes without a GPU, so why not make yourself a \
-            cup of tea in the meantime? (Or try the teticio/audio-diffusion-ddim-256 \
-                model which is faster.)",
-                    inputs=[
-                        gr.Dropdown(label="Model",
-                                    choices=[
-                                        "teticio/audio-diffusion-256",
-                                        "teticio/audio-diffusion-breaks-256",
-                                        "teticio/audio-diffusion-instrumental-hiphop-256",
-                                        "teticio/audio-diffusion-ddim-256"
-                                    ],
-                                    value="teticio/audio-diffusion-256")
+demo = gr.Interface(
+    fn=generate_spectrogram_audio_and_loop,
+    title="Audio Diffusion",
+    description="Generate audio using Huggingface diffusers.\
+        The models without 'latent' or 'ddim' give better results but take about \
+            20 minutes without a GPU.",
+    inputs=[
+        gr.Dropdown(label="Model",
+                    choices=[
+                        "teticio/audio-diffusion-256",
+                        "teticio/audio-diffusion-breaks-256",
+                        "teticio/audio-diffusion-instrumental-hiphop-256",
+                        "teticio/audio-diffusion-ddim-256",
+                        "teticio/latent-audio-diffusion-256",
+                        "teticio/latent-audio-diffusion-ddim-256"
                     ],
-                    outputs=[
-                        gr.Image(label="Mel spectrogram", image_mode="L"),
-                        gr.Audio(label="Audio"),
-                        gr.Audio(label="Loop"),
-                    ],
-                    allow_flagging="never")
+                    value="teticio/latent-audio-diffusion-ddim-256")
+    ],
+    outputs=[
+        gr.Image(label="Mel spectrogram", image_mode="L"),
+        gr.Audio(label="Audio"),
+        gr.Audio(label="Loop"),
+    ],
+    allow_flagging="never")
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()

streamlit_app.py

@@ -8,16 +8,19 @@ from audiodiffusion import AudioDiffusion
 
 if __name__ == "__main__":
     st.header("Audio Diffusion")
-    st.markdown("Generate audio using Huggingface diffusers.\
-        This takes about 20 minutes without a GPU, so why not make yourself a \
-            cup of tea in the meantime? (Or try the teticio/audio-diffusion-ddim-256 \
-                model which is faster.)")
+    st.markdown(
+        "Generate audio using Huggingface diffusers.\
+        The models without 'latent' or 'ddim' give better results but take about \
+            20 minutes without a GPU.", )
 
     model_id = st.selectbox("Model", [
         "teticio/audio-diffusion-256", "teticio/audio-diffusion-breaks-256",
         "teticio/audio-diffusion-instrumental-hiphop-256",
-        "teticio/audio-diffusion-ddim-256"
-    ])
+        "teticio/audio-diffusion-ddim-256",
+        "teticio/latent-audio-diffusion-256",
+        "teticio/latent-audio-diffusion-ddim-256"
+    ],
+                            index=5)
     audio_diffusion = AudioDiffusion(model_id=model_id)
 
     if st.button("Generate"):