Update app.py

app.py

@@ -94,3 +94,30 @@ if uploaded_files:
     n_candidates = st.slider("Number waveforms to generate", 1, 3, 3, 1)
 
     def score_waveforms(text, waveforms):
+    inputs = processor(text=text, audios=list(waveforms), return_tensors="pt", padding=True)
+    inputs = {key: inputs[key].to(device) for key in inputs}
+    with torch.no_grad():
+        logits_per_text = clap_model(**inputs).logits_per_text  # this is the audio-text similarity score
+        probs = logits_per_text.softmax(dim=-1)  # we can take the softmax to get the label probabilities
+        most_probable = torch.argmax(probs)  # and now select the most likely audio waveform
+    waveform = waveforms[most_probable]
+    return waveform
+
+if st.button("Générer de la musique"):
+    waveforms = pipe(
+        music_input,
+        audio_length_in_s=duration,
+        guidance_scale=guidance_scale,
+        num_inference_steps=100,
+        num_waveforms_per_prompt=n_candidates if n_candidates else 1,
+        generator=generator.manual_seed(int(seed)),
+    )["audios"]
+
+    if waveforms.shape[0] > 1:
+        waveform = score_waveforms(music_input, waveforms)
+    else:
+        waveform = waveforms[0]
+
+    # Afficher le lecteur audio
+    st.audio(waveform, format="audio/wav", sample_rate=16000)
+