Update app.py

app.py

@@ -1,18 +1,19 @@
-import gradio as gr
+from TTS.api import TTS
 from transformers import BlipProcessor, BlipForConditionalGeneration, AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
-from transformers import AutoTokenizer
+import torchaudio
+from torchaudio.transforms import Resample
+import torch
+import gradio as gr
+
+# Initialize TTS model from TTS library
+tts_model_path = "tts_models/multilingual/multi-dataset/xtts_v2"
+tts = TTS(tts_model_path, gpu=True)
 
 # Initialize Blip model for image captioning
 model_id = "dblasko/blip-dalle3-img2prompt"
 blip_model = BlipForConditionalGeneration.from_pretrained(model_id)
 blip_processor = BlipProcessor.from_pretrained(model_id)
 
-# Initialize TTS model from Hugging Face
-model_name = "tts-mozilla/tts-ljspeech-multilingual"
-tts_model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
-tts_tokenizer = AutoTokenizer.from_pretrained(model_name)
-tts = pipeline(task="text2speech", model=tts_model, tokenizer=tts_tokenizer)
-
 def generate_caption(image):
     # Generate caption from image using Blip model
     inputs = blip_processor(images=image, return_tensors="pt")
@@ -21,20 +22,30 @@ def generate_caption(image):
     generated_caption = blip_processor.batch_decode(generated_ids, skip_special_tokens=True, temperature=0.8, top_k=40, top_p=0.9)[0]
 
     # Use TTS model to convert generated caption to audio
-    audio_output = tts(generated_caption)
-    audio_output.save_to_path("generated_audio.mp3")
+    tts.tts_to_file(text=generated_caption,
+                    file_path="generated_audio.wav",
+                    speaker_wav="/path/to/target/speaker.wav",
+                    language="en")
+
+    # Resample the audio to match the expected sampling rate
+    waveform, sample_rate = torchaudio.load("generated_audio.wav")
+    resampler = Resample(orig_freq=sample_rate, new_freq=24_000)
+    waveform_resampled = resampler(waveform)
+
+    # Save the resampled audio
+    torchaudio.save("generated_audio_resampled.wav", waveform_resampled, 24_000)
 
-    return generated_caption, "generated_audio.mp3"
+    return generated_caption, "generated_audio_resampled.wav"
 
 # Create a Gradio interface with an image input, a textbox output, a button, and an audio player
 demo = gr.Interface(
-    fn=generate_caption, 
-    inputs=gr.Image(), 
+    fn=generate_caption,
+    inputs=gr.Image(),
     outputs=[
-        gr.Textbox(label="Generated caption"), 
-        gr.Button("Convert to Audio"), 
+        gr.Textbox(label="Generated caption"),
+        gr.Button("Convert to Audio"),
         gr.Audio(type="player", label="Generated Audio")
     ],
-    live=True  # ทำให้ Gradio ทำงานแบบไม่บล็อก
+    live=True
 )
 demo.launch(share=True)