Update app.py

add app changes to choose random voices references

app.py

@@ -73,8 +73,24 @@ _ = utils.load_checkpoint("checkpoints/freevc-mls.pth", freevc_mls, None)
 
 print("Loading WavLM for content...")
 cmodel = WavLMModel.from_pretrained("microsoft/wavlm-large").to(device)
- 
-def convert(model, src_mic,src_file, tgt):
+
+def get_random_wav_from_directory(directory, gender=None):
+    """
+    Get a random WAV file from a directory.
+    If gender is specified, it fetches a male or female WAV accordingly.
+    """
+    all_files = [f for f in os.listdir(directory) if f.endswith('.wav')]
+    
+    if gender == "male":
+        all_files = [f for f in all_files if "male" in f and "female" not in f]
+    elif gender == "female":
+        all_files = [f for f in all_files if "female" in f]
+        
+    return random.choice(all_files)
+
+
+
+def convert(model, src_mic,src_file, reference_option):
     """
     helper function which checks where source come from
     """
@@ -91,9 +107,16 @@ def convert(model, src_mic,src_file, tgt):
         logging.error("Source audio not provided")
         return
 
-    #if not tgt:
-    #    tgt="52_4703_000002.wav"
- 
+    if reference_option == "aléatoire":
+        tgt = get_random_wav_from_directory("mls_samples")
+    elif reference_option == "aléatoire (homme)":
+        tgt = get_random_wav_from_directory("mls_samples", "male")
+    elif reference_option == "aléatoire (femme)":
+        tgt = get_random_wav_from_directory("mls_samples", "female")
+    else:
+        logging.error("Option de référence non reconnue")
+        return
+    
     with torch.no_grad():
         # tgt
         wav_tgt, _ = librosa.load(tgt, sr=hps.data.sampling_rate)
@@ -139,8 +162,9 @@ def convert(model, src_mic,src_file, tgt):
 model = gr.Dropdown(choices=["FreeVC MLS","FreeVC (24kHz)"], value="FreeVC MLS",type="value", label="Model") 
 audio1_mic=gr.Audio(source="microphone", type="filepath", label='record your voice', optional=True)
 audio1_file = gr.inputs.Audio(type='filepath', label='or upload an audio file', optional=True)
-audio2 = gr.inputs.Audio(label="Reference Audio", type='filepath', optional=True)
-inputs = [model, audio1_mic, audio1_file, audio2]
+#audio2 = gr.inputs.Audio(label="Reference Audio", type='filepath', optional=True)
+reference_dropdown = gr.Dropdown(choices=["aléatoire", "aléatoire (homme)", "aléatoire (femme)"], value="aléatoire",label="Voix de référence")
+inputs = [model, audio1_mic, audio1_file, reference_dropdown]
 outputs = gr.outputs.Audio(label="Output Audio", type='filepath')
 
 title = "Démonstration d'Anonymisation de Voix"