Update app.py

app.py

@@ -9,6 +9,34 @@ import cv2
 import numpy as np
 from PIL import Image
 
+from parler_tts import ParlerTTSForConditionalGeneration
+from transformers import AutoTokenizer
+import soundfile as sf
+
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
+
+tts_model = ParlerTTSForConditionalGeneration.from_pretrained("CONCREE/Adia_TTS").to(device)
+tts_tokenizer = AutoTokenizer.from_pretrained("CONCREE/Adia_TTS")
+
+@spaces.GPU
+def tts(text):
+    output_wav_path = tempfile.mktemp(suffix=".wav")
+    # Description du style vocal
+    description = "A clear and educational voice, with a flow adapted to learning"
+    
+    # Génération
+    input_ids = tts_tokenizer(description, return_tensors="pt").input_ids.to(device)
+    prompt_ids = tts_tokenizer(text, return_tensors="pt").input_ids.to(device)
+    
+    audio = tts_model.generate(
+        input_ids=input_ids,
+        prompt_input_ids=prompt_ids,
+    )
+    sf.write(output_wav_path, audio.cpu().numpy().squeeze(), model.config.sampling_rate)
+
+    return output_wav_path
+
+
 def progress_bar_html(label: str) -> str:
     """
     Returns an HTML snippet for a thin progress bar with a label.
@@ -108,9 +136,8 @@ def model_inference(input_dict, history):
         buffer = ""
         for new_text in streamer:
             buffer += new_text
-            time.sleep(0.01)
+            time.sleep(0.001)
             yield buffer
-        return
 
     if len(files) > 1:
         images = [load_image(image) for image in files]
@@ -152,11 +179,14 @@ def model_inference(input_dict, history):
         time.sleep(0.01)
         yield buffer
 
+    return tts("Munul")
+
 demo = gr.ChatInterface(
     fn=model_inference,
     description="# oolel-vision-experimental `@video-infer for video understanding`**",
     fill_height=True,
     textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image", "video"], file_count="multiple"),
+    outputs=gr.Audio(label="Generated Speech")
     stop_btn="Stop Generation",
     multimodal=True,
     cache_examples=False,