Update app.py

app.py

@@ -47,10 +47,17 @@ doge_generation_config = GenerationConfig(
     repetition_penalty=1.0
 )
 
-# Load and trim speaker embedding for SpeechT5 (must be size 600)
+# Load and normalize speaker embedding to 600 dimensions
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-speaker_embedding = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)[:, :600]
+speaker_embedding = None
+for i in range(len(embeddings_dataset)):
+    vec = embeddings_dataset[i]["xvector"]
+    if len(vec) >= 600:
+        speaker_embedding = torch.tensor(vec[:600]).unsqueeze(0)
+        break
 
+if speaker_embedding is None:
+    raise ValueError("No suitable speaker embedding of at least 600 dimensions found.")
 
 def process_image(image):
     try:
@@ -109,7 +116,7 @@ iface = gr.Interface(
         gr.Textbox(label="Extracted Text (OCR)"),
         gr.Textbox(label="Generated Context")
     ],
-    title="SeeSay Contextualizer with Doge-320M",
+    title="SeeSay",
     description="Upload an image to generate a caption, extract text (OCR), generate context using Doge, and turn it into speech using SpeechT5."
 )