Update app.py

app.py

@@ -1,8 +1,6 @@
-# import part
+# import part - only using the two requested imports
 import streamlit as st
 from transformers import pipeline
-import os
-import tempfile
 
 # function part
 # img2text
@@ -11,18 +9,18 @@ def img2text(image_path):
     text = image_to_text(image_path)[0]["generated_text"]
     return text
 
-# text2story
+# text2story - IMPROVED to end naturally
 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, "
+    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 the story
     story_result = generator(
         prompt,
-        max_length=150,
+        max_length=250,  # Increased to allow for a complete story
         num_return_sequences=1,
         temperature=0.7,
         top_k=50,
@@ -34,25 +32,34 @@ def text2story(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
+    # Find a natural ending point (end of sentence) before 100 words
     words = story_text.split()
     if len(words) > 100:
-        # Simply truncate to 100 words
-        story_text = " ".join(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 - REVISED to correctly handle the audio output
+# text2audio - Using HelpingAI-TTS-v1 model
 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"]
+        # 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
@@ -63,71 +70,18 @@ def text2audio(story_text):
             else:
                 story_text = story_text[:max_chars]
         
-        # Generate speech with correct parameters
-        speech = synthesizer(
-            text=story_text,
-            forward_params={"speaker_embeddings": speaker_embeddings}
-        )
+        # 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()
-        
-        # 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
+        # Get output information
+        st.write(f"Speech output keys: {list(speech.keys())}")
         
-        return temp_filename
+        return speech
         
     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")
@@ -137,12 +91,12 @@ 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)
+    # Create a temporary file in memory from the uploaded file
+    image_bytes = uploaded_file.getvalue()
     
     # Stage 1: Image to Text
     st.text('Processing img2text...')
-    caption = img2text(image_path)
+    caption = img2text(image_bytes)  # Pass bytes directly to pipeline
     st.write(caption)
     
     # Stage 2: Text to Story
@@ -152,19 +106,35 @@ if uploaded_file is not None:
     
     # Stage 3: Story to Audio data
     st.text('Generating audio data...')
-    audio_file = text2audio(story)
+    speech_output = 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)
+        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.")
-    
-    # Clean up the temporary files
-    try:
-        os.remove(image_path)
-        # Don't delete audio file immediately as it might still be playing
-    except:
-        pass
+            st.error("Audio generation failed. Please try again.")