Spaces:
Sleeping
Sleeping
| # import part | |
| import streamlit as st | |
| from transformers import pipeline | |
| import os | |
| import numpy as np | |
| import io | |
| # 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 use a simpler approach without scipy | |
| def text2audio(story_text): | |
| try: | |
| # Use the facebook/mms-tts-eng model with fewer features | |
| synthesizer = pipeline("text-to-speech", model="facebook/mms-tts-eng") | |
| # For simplicity, we'll limit the text length to avoid timeouts | |
| # If text is too long, truncate it to a reasonable length (500 chars ~ 100 words) | |
| max_length = 500 | |
| if len(story_text) > max_length: | |
| last_period = story_text[:max_length].rfind('.') | |
| if last_period > 0: | |
| story_text = story_text[:last_period + 1] | |
| else: | |
| story_text = story_text[:max_length] | |
| # Generate speech | |
| speech = synthesizer(story_text) | |
| # Save the audio to a file instead of using in-memory processing | |
| # This avoids needing scipy | |
| temp_audio_path = "temp_audio.wav" | |
| # Convert numpy array to bytes and save | |
| with open(temp_audio_path, "wb") as f: | |
| # Assuming the audio is in the right format already | |
| np.save(f, speech["audio"]) | |
| # Read the file back | |
| with open(temp_audio_path, "rb") as f: | |
| audio_data = f.read() | |
| # Clean up | |
| try: | |
| os.remove(temp_audio_path) | |
| except: | |
| pass | |
| return { | |
| "audio": audio_data, | |
| "sampling_rate": speech["sampling_rate"] | |
| } | |
| except Exception as e: | |
| st.error(f"Error generating audio: {str(e)}") | |
| # No fallback - just return None | |
| 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_data = text2audio(story) | |
| # Play button | |
| if st.button("Play Audio"): | |
| if audio_data: | |
| st.audio( | |
| audio_data["audio"], | |
| format="audio/wav", | |
| start_time=0, | |
| sample_rate=audio_data["sampling_rate"] | |
| ) | |
| else: | |
| st.error("Failed to generate audio. Please try again.") | |
| # Clean up the temporary file | |
| try: | |
| os.remove(image_path) | |
| except: | |
| pass |