Update app.py

app.py

@@ -2,7 +2,6 @@
 # Step 0: Import required libraries
 ##########################################
 import streamlit as st  # For building the web application interface
-import soundfile as sf  # For saving audio as .wav files
 from transformers import (
     pipeline,
     SpeechT5Processor,
@@ -12,6 +11,9 @@ from transformers import (
     AutoTokenizer
 )  # For sentiment analysis, text-to-speech, and text generation
 from datasets import load_dataset  # For loading datasets (e.g., speaker embeddings)
+import torch  # For tensor operations
+import soundfile as sf  # For saving audio as .wav files
+import sentencepiece  # Required by SpeechT5Processor for tokenization
 
 
 ##########################################
@@ -27,7 +29,7 @@ st.markdown(
 st.markdown(
     "<h3 style='text-align: center; color: #5D6D7E; font-style: italic;'>I'm listening to you, my friend</h3>",
     unsafe_allow_html=True
-)  # Use HTML to add a friendly and soft-styled subtitle
+)  # Use HTML for a friendly and soft-styled subtitle
 
 # Add a well-designed text area for user input
 text = st.text_area(
@@ -48,17 +50,17 @@ def analyze_dominant_emotion(user_review):
         "text-classification", 
         model="Thea231/jhartmann_emotion_finetuning", 
         return_all_scores=True
-    )  # Load the fine-tuned text classification model from Hugging Face
+    )  # Load the fine-tuned text classification model
 
     emotion_results = emotion_classifier(user_review)[0]  # Perform sentiment analysis on the input text
-    dominant_emotion = max(emotion_results, key=lambda x: x['score'])  # Identify the emotion with the highest confidence score
+    dominant_emotion = max(emotion_results, key=lambda x: x['score'])  # Identify the emotion with the highest confidence
     return dominant_emotion  # Return the dominant emotion (label and score)
 
-
 ##########################################
 # Step 2: Response Generation Function
 ##########################################
 
+
 def response_gen(user_review):
     """
     Generate a concise and logical response based on the sentiment of the user's comment.
@@ -132,28 +134,28 @@ def response_gen(user_review):
         )
     }
     
-    # Select the appropriate prompt based on the user's emotion or default to neutral
+    # Select the appropriate prompt based on the user's emotion
     prompt = emotion_prompts.get(
         emotion_label, 
         f"Neutral feedback: '{user_review}'\n\nWrite a professional and concise response (50-200 words max).\n\nResponse:"
     )
 
     # Load the tokenizer and language model for response generation
-    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-0.5B")  # Load tokenizer for processing text inputs
+    tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-0.5B")  # Load tokenizer for text processing
     model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen1.5-0.5B")  # Load language model for text generation
 
     inputs = tokenizer(prompt, return_tensors="pt")  # Tokenize the input prompt
     outputs = model.generate(
         **inputs,
-        max_new_tokens=300,  # Set an upper limit on token generation to ensure concise output
-        min_length=75,  # Set a minimum length to ensure the response is complete
+        max_new_tokens=300,  # Limit generated tokens to ensure concise responses
+        min_length=75,  # Ensure the generated response is logical and complete
         no_repeat_ngram_size=2,  # Avoid repetitive phrases
-        temperature=0.7  # Add randomness for more natural responses
+        temperature=0.7  # Add randomness for natural-sounding responses
     )
 
-    # Decode the generated response back into readable text
+    # Decode the generated response back into text
     response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    print(f"{response}")  # Print the response for debugging
+    print(f"Generated response: {response}")  # Debugging: print the response
     return response  # Return the generated response
 
 ##########################################
@@ -163,43 +165,43 @@ def sound_gen(response):
     """
     Convert the generated response to speech and save it as a .wav file.
     """
-    processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")  # Pre-trained processor for TTS
-    model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts")  # Pre-trained TTS model
-    vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")  # Vocoder for generating waveforms
+    processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")  # Load processor for TTS
+    model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts")  # Load pre-trained TTS model
+    vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")  # Load vocoder for waveform generation
 
-    # Load a neutral female voice embedding from a pre-trained dataset
+    # Load neutral female voice embedding from dataset
     embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")  # Load speaker embeddings
     speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)  # Use a default speaker embedding
 
-    # Process the input text and create a speech spectrogram
+    # Process the input text and generate a spectrogram
     inputs = processor(text=response, return_tensors="pt")
     spectrogram = model.generate_speech(inputs["input_ids"], speaker_embeddings)
 
-    # Convert the spectrogram into an audio waveform using the vocoder
+    # Use vocoder to convert the spectrogram into a waveform
     with torch.no_grad():
         speech = vocoder(spectrogram)
 
-    # Save the audio as a .wav file
+    # Save the audio file as .wav
     sf.write("customer_service_response.wav", speech.numpy(), samplerate=16000)
 
-    # Embed an auto-playing audio player in the web app
-    st.audio("customer_service_response.wav", start_time=0)  # Allow audio playback with autoplay feature
+    # Create an auto-playing audio player in Streamlit
+    st.audio("customer_service_response.wav", start_time=0)  # Enable audio playback with autoplay
 
 ##########################################
 # Main Function
 ##########################################
 def main():
     """
-    Main function to orchestrate the workflow of sentiment analysis, response generation, and text-to-speech.
+    Main function to handle sentiment analysis, response generation, and text-to-speech functionalities.
     """
     if text:  # Check if the user has entered a comment
-        response = response_gen(text)  # Generate a logical and concise response
+        response = response_gen(text)  # Generate a concise and logical response
         st.markdown(
             f"<p style='color:#2ECC71; font-size:20px;'>{response}</p>", 
             unsafe_allow_html=True
-        )  # Display the generated response in a cute, styled font
-        sound_gen(response)  # Convert the response to speech and make it available for playback
+        )  # Display the response in a styled font
+        sound_gen(response)  # Convert the response to speech and play it
 
-# Run the main function when the script is executed
+# Execute the main function
 if __name__ == "__main__":
     main()