Create app.py

app.py

@@ -0,0 +1,93 @@
+import whisper  # Import whisper from OpenAI
+import streamlit as st
+import librosa
+import numpy as np
+import matplotlib.pyplot as plt
+from pydub import AudioSegment
+from transformers import pipeline
+import os
+
+# Load pre-trained sentiment analysis model
+sentiment_analyzer = pipeline("sentiment-analysis")
+
+# Streamlit UI
+st.title("🎤 Audio Sentiment Analysis")
+st.write("Upload multiple MP3 files to analyze sentiment and tone.")
+
+# Upload multiple audio files
+uploaded_files = st.file_uploader("Choose MP3 files", type=["mp3"], accept_multiple_files=True)
+
+# Function to process audio, get sentiment, and transcribe to text using Whisper
+def analyze_audio(file_path):
+    # Convert MP3 to WAV
+    audio = AudioSegment.from_mp3(file_path)
+    wav_path = file_path.replace(".mp3", ".wav")
+    audio.export(wav_path, format="wav")
+
+    # Load audio
+    y, sr = librosa.load(wav_path, sr=None)
+    
+    # Extract MFCCs (Mel-frequency cepstral coefficients)
+    mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13)
+    mfccs_mean = np.mean(mfccs, axis=1)  # Take mean across time axis
+
+    # Perform speech-to-text transcription using Whisper
+    model = whisper.load_model("base")  # Load Whisper model (change model size as needed)
+    result = model.transcribe(wav_path)
+    transcription = result['text']
+    
+    # Perform sentiment analysis on the transcription
+    sentiment_result = sentiment_analyzer(transcription) if transcription else [{"label": "NEGATIVE", "score": 0.5}]
+    
+    # Remove WAV file after processing
+    os.remove(wav_path)
+    
+    return sentiment_result[0], mfccs_mean, mfccs, y, sr, transcription
+
+# Process and plot if files are uploaded
+if uploaded_files:
+    # Create a directory to store temporary files
+    os.makedirs("temp", exist_ok=True)
+
+    # Prepare a single plot
+    fig, ax = plt.subplots(figsize=(10, 6))
+
+    for uploaded_file in uploaded_files:
+        # Save the uploaded file
+        file_path = f"temp/{uploaded_file.name}"
+        with open(file_path, "wb") as f:
+            f.write(uploaded_file.getbuffer())
+
+        # Analyze sentiment, extract features, and get transcription
+        sentiment, mfccs_mean, mfccs, audio_data, sample_rate, transcription = analyze_audio(file_path)
+
+        # Display sentiment and transcription result
+        st.subheader(f"📊 Sentiment Analysis Result for {uploaded_file.name}")
+        st.write(f"**Transcription:** {transcription}")
+        st.write(f"**Sentiment:** {sentiment['label']}")
+        st.write(f"**Confidence:** {sentiment['score']:.2f}")
+
+        # Normalize sentiment score to range from 0 to 1
+        sentiment_score = sentiment['score'] if sentiment['label'] == 'POSITIVE' else 1 - sentiment['score']
+
+        # Plotting both curves in a single plot
+        ax.plot(np.linspace(0, len(audio_data) / sample_rate, len(audio_data)), [sentiment_score] * len(audio_data), label="Agent's Tone", linestyle='-', color='b')
+        
+        # Ensure that MFCCs are being averaged across time correctly
+        # Take the mean along the time axis (axis=1) to get one value per MFCC coefficient
+        mfccs_mean = np.mean(mfccs, axis=1)
+
+        # Plot Patient's Tone Curve (MFCC mean value per coefficient)
+        ax.plot(np.linspace(0, len(mfccs_mean), len(mfccs_mean)), mfccs_mean, label="Patient's Tone Curve (MFCCs)", linestyle='--', color='r')
+
+    # Customize plot
+    ax.set_xlabel("Time (seconds) / MFCC Coefficients")
+    ax.set_ylabel("Score / Mean MFCC Value")
+    ax.set_title("Agent Tone & Patient Tone Curve")
+    ax.legend()
+
+    # Show plot in Streamlit
+    st.pyplot(fig)
+
+    # Clean up temp files
+    os.remove(file_path)