Update app.py

app.py

@@ -1,29 +1,26 @@
 # 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
+# img2text
+def img2text(image_path):
     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"]
+    text = image_to_text(image_path)[0]["generated_text"]
     return text
 
-# text2story - Using the original model but with optimized parameters
+# text2story - IMPROVED to end naturally
 def text2story(text):
-    # Using the specified TinyLlama model
+    # 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 brief children's story based on this: {text}. Once upon a time, "
+    prompt = f"Write a fun children's story based on this: {text}. The story should be short and end naturally with a conclusion. Once upon a time, "
     
-    # Generate with more constrained parameters for speed
+    # Generate the story
     story_result = generator(
         prompt,
-        max_new_tokens=150,  # Use max_new_tokens instead of max_length for efficiency
+        max_length=250,  # Increased to allow for a complete story
         num_return_sequences=1,
         temperature=0.7,
         top_k=50,
@@ -58,7 +55,7 @@ def text2story(text):
     
     return story_text
 
-# text2audio - Using HelpingAI-TTS-v1 model
+# text2audio - Simplified without numpy/scipy
 def text2audio(story_text):
     try:
         # Use the HelpingAI TTS model as requested
@@ -74,14 +71,32 @@ def text2audio(story_text):
                 story_text = story_text[:max_chars]
         
         # Generate speech
+        st.write("Generating audio...")
         speech = synthesizer(story_text)
+        st.write(f"Speech output keys: {list(speech.keys())}")
         
+        # We'll pass the audio data directly to Streamlit instead of saving to a file
+        # This works because Streamlit's st.audio() can take raw audio data
         return speech
         
     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")
@@ -91,28 +106,22 @@ 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)
+    # Save the image temporarily
+    image_path = save_uploaded_image(uploaded_file)
     
     # Stage 1: Image to Text
-    with st.spinner('Processing image caption...'):
-        caption = img2text(image)
-        progress_bar.progress(33)
-    st.write(f"**Image caption:** {caption}")
+    st.text('Processing img2text...')
+    caption = img2text(image_path)
+    st.write(caption)
     
     # Stage 2: Text to Story
-    with st.spinner('Creating story...'):
-        story = text2story(caption)
-        progress_bar.progress(66)
-    st.write(f"**Story:** {story}")
+    st.text('Generating a story...')
+    story = text2story(caption)
+    st.write(story)
     
     # Stage 3: Story to Audio data
-    with st.spinner('Generating audio...'):
-        speech_output = text2audio(story)
-        progress_bar.progress(100)
+    st.text('Generating audio data...')
+    speech_output = text2audio(story)
     
     # Play button
     if st.button("Play Audio"):
@@ -126,21 +135,14 @@ if uploaded_file is not None:
                 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())}")
+                    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.")
+            st.error("Audio generation failed. Please try again.")
+    
+    # Clean up the temporary files
+    try:
+        os.remove(image_path)
+    except:
+        pass