Spaces:
Sleeping
Sleeping
| # 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 correctly handle the audio output | |
| def text2audio(story_text): | |
| try: | |
| # Use a different TTS model that works reliably with pipeline | |
| synthesizer = pipeline("text-to-speech", model="microsoft/speecht5_tts") | |
| # Additional input required for this model | |
| speaker_embeddings = pipeline( | |
| "audio-classification", | |
| model="microsoft/speecht5_speaker_embeddings" | |
| )("some_audio_file.mp3")["logits"] | |
| # 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 with correct parameters | |
| speech = synthesizer( | |
| text=story_text, | |
| forward_params={"speaker_embeddings": speaker_embeddings} | |
| ) | |
| # Create a temporary WAV file | |
| temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.wav') | |
| temp_filename = temp_file.name | |
| temp_file.close() | |
| # Display the structure of the speech output for debugging | |
| st.write(f"Speech output keys: {speech.keys()}") | |
| # Save the audio data to the temporary file | |
| # Different models have different output formats, we'll try common keys | |
| if 'audio' in speech: | |
| # Convert numpy array to WAV file | |
| try: | |
| import scipy.io.wavfile as wavfile | |
| wavfile.write(temp_filename, speech['sampling_rate'], speech['audio']) | |
| except ImportError: | |
| # If scipy is not available, try raw writing | |
| with open(temp_filename, 'wb') as f: | |
| # Convert numpy array to bytes in a simple way | |
| if isinstance(speech['audio'], np.ndarray): | |
| audio_bytes = speech['audio'].tobytes() | |
| f.write(audio_bytes) | |
| else: | |
| f.write(speech['audio']) | |
| elif 'numpy_array' in speech: | |
| with open(temp_filename, 'wb') as f: | |
| f.write(speech['numpy_array'].tobytes()) | |
| else: | |
| # Fallback: try to write whatever is available | |
| with open(temp_filename, 'wb') as f: | |
| # Just write the first value that seems like it could be audio data | |
| for key, value in speech.items(): | |
| if isinstance(value, (bytes, bytearray)) or ( | |
| isinstance(value, np.ndarray) and value.size > 1000): | |
| if isinstance(value, np.ndarray): | |
| f.write(value.tobytes()) | |
| else: | |
| f.write(value) | |
| break | |
| return temp_filename | |
| except Exception as e: | |
| st.error(f"Error generating audio: {str(e)}") | |
| # Print all available keys for debugging | |
| 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 |