Spaces:
Sleeping
Sleeping
| # import part | |
| import streamlit as st | |
| from transformers import pipeline | |
| from PIL import Image | |
| # function part | |
| # img2text - Using the original model | |
| def img2text(image): | |
| # Use the specified model but with optimized parameters | |
| image_to_text = pipeline("image-to-text", model="sooh-j/blip-image-captioning-base") | |
| # Limiting the output length for speed | |
| text = image_to_text(image, max_new_tokens=30)[0]["generated_text"] | |
| return text | |
| # text2story - Using the original model but with optimized parameters | |
| def text2story(text): | |
| # Using the specified TinyLlama model | |
| generator = pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0") | |
| # Create a prompt for the story generation | |
| prompt = f"Write a brief children's story based on this: {text}. Once upon a time, " | |
| # Generate with more constrained parameters for speed | |
| story_result = generator( | |
| prompt, | |
| max_new_tokens=150, # Use max_new_tokens instead of max_length for efficiency | |
| 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, ") | |
| # Find a natural ending point (end of sentence) before 100 words | |
| words = story_text.split() | |
| if len(words) > 100: | |
| # Join the first 100 words | |
| shortened_text = " ".join(words[:100]) | |
| # Find the last complete sentence | |
| last_period = shortened_text.rfind('.') | |
| last_question = shortened_text.rfind('?') | |
| last_exclamation = shortened_text.rfind('!') | |
| # Find the last sentence ending punctuation | |
| last_end = max(last_period, last_question, last_exclamation) | |
| if last_end > 0: | |
| # Truncate at the end of the last complete sentence | |
| story_text = shortened_text[:last_end + 1] | |
| else: | |
| # If no sentence ending found, just use the shortened text | |
| story_text = shortened_text | |
| return story_text | |
| # text2audio - Using HelpingAI-TTS-v1 model | |
| def text2audio(story_text): | |
| try: | |
| # Use the HelpingAI TTS model as requested | |
| synthesizer = pipeline("text-to-speech", model="HelpingAI/HelpingAI-TTS-v1") | |
| # 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) | |
| return speech | |
| except Exception as e: | |
| st.error(f"Error generating audio: {str(e)}") | |
| return None | |
| # 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) | |
| # Convert the file to a PIL Image | |
| image = Image.open(uploaded_file) | |
| # Progress indicator | |
| progress_bar = st.progress(0) | |
| # Stage 1: Image to Text | |
| with st.spinner('Processing image caption...'): | |
| caption = img2text(image) | |
| progress_bar.progress(33) | |
| st.write(f"**Image caption:** {caption}") | |
| # Stage 2: Text to Story | |
| with st.spinner('Creating story...'): | |
| story = text2story(caption) | |
| progress_bar.progress(66) | |
| st.write(f"**Story:** {story}") | |
| # Stage 3: Story to Audio data | |
| with st.spinner('Generating audio...'): | |
| speech_output = text2audio(story) | |
| progress_bar.progress(100) | |
| # Play button | |
| if st.button("Play Audio"): | |
| if speech_output is not None: | |
| # Try to play the audio directly | |
| try: | |
| if 'audio' in speech_output and 'sampling_rate' in speech_output: | |
| st.audio(speech_output['audio'], sample_rate=speech_output['sampling_rate']) | |
| elif 'audio_array' in speech_output and 'sampling_rate' in speech_output: | |
| st.audio(speech_output['audio_array'], sample_rate=speech_output['sampling_rate']) | |
| elif 'waveform' in speech_output and 'sample_rate' in speech_output: | |
| st.audio(speech_output['waveform'], sample_rate=speech_output['sample_rate']) | |
| else: | |
| # Try the first array-like value as audio data | |
| for key, value in speech_output.items(): | |
| if hasattr(value, '__len__') and len(value) > 1000: | |
| if 'rate' in speech_output: | |
| st.audio(value, sample_rate=speech_output['rate']) | |
| elif 'sample_rate' in speech_output: | |
| st.audio(value, sample_rate=speech_output['sample_rate']) | |
| elif 'sampling_rate' in speech_output: | |
| st.audio(value, sample_rate=speech_output['sampling_rate']) | |
| else: | |
| st.audio(value, sample_rate=24000) # Default sample rate | |
| break | |
| else: | |
| st.error(f"Could not find compatible audio format in: {list(speech_output.keys())}") | |
| except Exception as e: | |
| st.error(f"Error playing audio: {str(e)}") | |
| else: | |
| st.error("Audio generation failed. Please try again.") |