app.py

# import part
import streamlit as st
from transformers import pipeline
import os
import tempfile

# function part
# img2text
def img2text(image_path):
    image_to_text = pipeline("image-to-text", model="sooh-j/blip-image-captioning-base")
    text = image_to_text(image_path)[0]["generated_text"]
    return text

# text2story
def text2story(text):
    # Using a smaller text generation model
    generator = pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0")
    
    # Create a prompt for the story generation
    prompt = f"Write a fun children's story based on this: {text}. Once upon a time, "
    
    # Generate the story
    story_result = generator(
        prompt,
        max_length=150,
        num_return_sequences=1,
        temperature=0.7,
        top_k=50,
        top_p=0.95,
        do_sample=True
    )
   
    # Extract the generated text
    story_text = story_result[0]['generated_text']
    story_text = story_text.replace(prompt, "Once upon a time, ")
    
    # Make sure the story is at least 100 words
    words = story_text.split()
    if len(words) > 100:
        # Simply truncate to 100 words
        story_text = " ".join(words[:100])
    
    return story_text

# text2audio - REVISED to handle audio format correctly
# text2audio - REVISED with proper audio field handling
def text2audio(story_text):
    try:
        # Use the MeloTTS model which has better audio quality
        synthesizer = pipeline("text-to-speech", model="myshell-ai/MeloTTS-English")
        
        # Limit text length to avoid timeouts
        max_chars = 500
        if len(story_text) > max_chars:
            last_period = story_text[:max_chars].rfind('.')
            if last_period > 0:
                story_text = story_text[:last_period + 1]
            else:
                story_text = story_text[:max_chars]
        
        # Generate speech
        speech = synthesizer(story_text)
        
        # Create a temporary WAV file
        temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.wav')
        temp_filename = temp_file.name
        temp_file.close()
        
        # Debug: Show what keys are available in the speech output
        st.write(f"Speech output keys: {list(speech.keys())}")
        
        # Write the audio data to the temporary file - MeloTTS should have audio and sampling_rate
        if 'audio' in speech and 'sampling_rate' in speech:
            # Convert numpy array to WAV file
            scipy.io.wavfile.write(
                temp_filename, 
                speech['sampling_rate'], 
                speech['audio'].astype(np.float32)
            )
            st.write("Audio successfully written to file")
        else:
            raise ValueError(f"Expected 'audio' and 'sampling_rate' in output, but got: {list(speech.keys())}")
        
        return temp_filename
        
    except Exception as e:
        st.error(f"Error generating audio: {str(e)}")
        import traceback
        st.error(traceback.format_exc())
        return None

# Function to save temporary image file
def save_uploaded_image(uploaded_file):
    if not os.path.exists("temp"):
        os.makedirs("temp")
    
    image_path = os.path.join("temp", uploaded_file.name)
    
    with open(image_path, "wb") as f:
        f.write(uploaded_file.getvalue())
    
    return image_path

# main part
st.set_page_config(page_title="Your Image to Audio Story", page_icon="🦜")
st.header("Turn Your Image to Audio Story")
uploaded_file = st.file_uploader("Select an Image...")

if uploaded_file is not None:
    # Display the uploaded image
    st.image(uploaded_file, caption="Uploaded Image", use_container_width=True)
    
    # Save the image temporarily
    image_path = save_uploaded_image(uploaded_file)
    
    # Stage 1: Image to Text
    st.text('Processing img2text...')
    caption = img2text(image_path)
    st.write(caption)
    
    # Stage 2: Text to Story
    st.text('Generating a story...')
    story = text2story(caption)
    st.write(story)
    
    # Stage 3: Story to Audio data
    st.text('Generating audio data...')
    audio_file = text2audio(story)
    
    # Play button
    if st.button("Play Audio"):
        if audio_file and os.path.exists(audio_file):
            # Play the audio file
            st.audio(audio_file)
        else:
            st.error("Audio generation failed. Please try again.")
    
    # Clean up the temporary files
    try:
        os.remove(image_path)
        # Don't delete audio file immediately as it might still be playing
    except:
        pass