Update app.py

app.py

@@ -23,6 +23,47 @@ st.write("Upload an MP3 file to analyze its sentiment and audio features.")
 # Upload audio file
 uploaded_file = st.file_uploader("Choose an MP3 file", type=["mp3"])
 
+# Function to analyze sentiment over time
+def analyze_sentiment_over_time(y, sr, chunk_duration=10):
+    """
+    Splits audio into chunks, transcribes, and performs sentiment analysis.
+
+    Args:
+        y: The audio waveform.
+        sr: Sample rate.
+        chunk_duration: Duration of each chunk in seconds (default: 10 sec).
+
+    Returns:
+        A list of sentiment labels over time.
+    """
+    chunk_length = chunk_duration * sr  # Convert chunk duration to samples
+    total_chunks = len(y) // chunk_length  # Number of chunks
+    sentiment_labels = []
+
+    for i in range(total_chunks):
+        start_sample = i * chunk_length
+        end_sample = start_sample + chunk_length
+        chunk_audio = y[start_sample:end_sample]
+
+        # Convert chunk to WAV
+        temp_wav_path = f"temp_chunk_{i}.wav"
+        librosa.output.write_wav(temp_wav_path, chunk_audio, sr)
+
+        # Transcribe with Whisper
+        result = whisper_model.transcribe(temp_wav_path)
+        os.remove(temp_wav_path)  # Remove temporary WAV file
+        
+        transcribed_text = result["text"]
+        
+        # Run sentiment analysis on transcribed text (if available)
+        if transcribed_text.strip():
+            sentiment_result = sentiment_analyzer(transcribed_text)
+            sentiment_labels.append(sentiment_result[0]["label"])
+        else:
+            sentiment_labels.append("NEUTRAL")
+
+    return sentiment_labels
+
 # Function to process audio and get sentiment
 def analyze_audio(file_path):
     # Convert MP3 to WAV
@@ -35,15 +76,20 @@ def analyze_audio(file_path):
     
     # 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
+    mfccs_mean = np.mean(mfccs, axis=1)
 
-    # Dummy text for sentiment analysis (replace with actual text from speech-to-text if needed)
-    sentiment_result = sentiment_analyzer("This is a placeholder for sentiment analysis based on audio!")
+    # Transcribe audio
+    result = whisper_model.transcribe(wav_path)
+    transcribed_text = result["text"]
 
-    # Remove WAV file after processing
-    os.remove(wav_path)
-    
-    return y, sr, sentiment_result[0], mfccs, mfccs_mean
+    # Run sentiment analysis
+    if transcribed_text.strip():
+        sentiment_result = sentiment_analyzer(transcribed_text)
+    else:
+        sentiment_result = [{"label": "NEUTRAL", "score": 0.0}]
+
+    os.remove(wav_path)  # Remove WAV file after processing
+    return y, sr, sentiment_result[0], mfccs, mfccs_mean, transcribed_text
 
 # Function to extract words from audio using Whisper
 def extract_words_from_audio(file_path):
@@ -55,123 +101,57 @@ def extract_words_from_audio(file_path):
     # Transcribe audio using Whisper
     result = whisper_model.transcribe(wav_path, word_timestamps=True)
 
-    # Extract words and their timestamps
+    # Extract words and timestamps
     words = []
     for segment in result['segments']:
         for word_info in segment['words']:
-            word = word_info['word']
-            start_time = word_info['start']
-            end_time = word_info['end']
-            words.append({"word": word, "start_time": start_time, "end_time": end_time})
+            words.append({"word": word_info['word'], "start_time": word_info['start'], "end_time": word_info['end']})
 
-    # Remove WAV file after processing
     os.remove(wav_path)
-    
     return words, result['text']
 
 # Process and plot if a file is uploaded
 if uploaded_file:
     file_path = f"temp/{uploaded_file.name}"
+    os.makedirs("temp", exist_ok=True)  # Ensure temp directory exists
 
-    # Ensure temp directory exists
-    os.makedirs("temp", exist_ok=True)
-
-    # Save the uploaded file
     with open(file_path, "wb") as f:
         f.write(uploaded_file.getbuffer())
 
-    # Analyze sentiment and extract features
-    y, sr, sentiment, mfccs, mfccs_mean = analyze_audio(file_path)
+    # Analyze sentiment & extract features
+    y, sr, sentiment, mfccs, mfccs_mean, transcribed_text = analyze_audio(file_path)
 
-    # Extract words from audio using Whisper
-    words_from_audio, transcribed_text = extract_words_from_audio(file_path)
+    # Extract words from audio
+    words_from_audio, _ = extract_words_from_audio(file_path)
 
-    # Display words from audio in two categories: good and negative words
-    good_words = [word['word'] for word in words_from_audio if word['word'].lower() in ['good', 'excellent', 'positive', 'great', 'happy', 'success']]
-    negative_words = [word['word'] for word in words_from_audio if word['word'].lower() in ['bad', 'negative', 'poor', 'angry', 'sad', 'failure']]
+    # Categorize words
+    good_words = [w['word'] for w in words_from_audio if w['word'].lower() in ['good', 'excellent', 'positive', 'great', 'happy', 'success']]
+    negative_words = [w['word'] for w in words_from_audio if w['word'].lower() in ['bad', 'negative', 'poor', 'angry', 'sad', 'failure']]
 
-    # Determine color based on sentiment
+    # Determine sentiment color
     sentiment_label = sentiment['label']
     sentiment_color = "green" if sentiment_label == "POSITIVE" else "red"
 
-    # Display sentiment result
+    # Display sentiment
     st.subheader("📊 Sentiment Analysis Result")
     st.markdown(f"**Sentiment:** <span style='color:{sentiment_color}; font-size:20px;'>{sentiment_label}</span>", unsafe_allow_html=True)
     st.write(f"**Confidence:** {sentiment['score']:.2f}")
 
-    # **Create a 2x3 Grid Layout for Plots with Larger Size and Adjustments**
-    fig, axes = plt.subplots(2, 3, figsize=(30, 18), dpi=300)  # Increased size and DPI for better quality and larger plots
-    axes = axes.flatten()  # Flatten the 2x3 grid to make it easier to index
-
-    # **Plot 1: Line Plot of Average MFCCs**
-    axes[0].plot(range(1, len(mfccs_mean) + 1), mfccs_mean, marker='o', color='b', label="Average MFCCs")
-    axes[0].set_xlabel("MFCC Coefficients")
-    axes[0].set_ylabel("Mean Value")
-    axes[0].set_title("Average MFCCs Across Time")
-    axes[0].legend()
-
-    # **Plot 2: Audio Waveform with Sentiment Indicator**
-    librosa.display.waveshow(y, sr=sr, ax=axes[1], alpha=0.5, color=sentiment_color)
-    axes[1].set_xlabel("Time (s)")
-    axes[1].set_ylabel("Amplitude")
-    axes[1].set_title(f"Waveform of Audio ({'Good' if sentiment_label == 'POSITIVE' else 'Bad'})")
-
-    # **Plot 3: MFCC Heatmap**
-    sns.heatmap(mfccs, ax=axes[2], cmap="coolwarm", yticklabels=[f"MFCC {i}" for i in range(1, 14)])
-    axes[2].set_xlabel("Time Frames")
-    axes[2].set_ylabel("MFCC Coefficients")
-    axes[2].set_title(f"MFCC Feature Heatmap ({'Good' if sentiment_label == 'POSITIVE' else 'Bad'})")
-
-    # **Plot 4: Spectrogram**
-    S = librosa.feature.melspectrogram(y=y, sr=sr)
-    S_dB = librosa.power_to_db(S, ref=np.max)
-    img = librosa.display.specshow(S_dB, sr=sr, x_axis="time", y_axis="mel", ax=axes[3])
-    fig.colorbar(img, ax=axes[3], format="%+2.0f dB")
-    axes[3].set_title(f"Mel Spectrogram ({'Good' if sentiment_label == 'POSITIVE' else 'Bad'})")
-
-    # **Plot 5: Sentiment Trend Over Time (10-sec intervals)**
+    # Analyze sentiment over time
     sentiment_scores = analyze_sentiment_over_time(y, sr)
-    sentiment_labels = list(set(sentiment_scores))  # Unique sentiment labels
-    
-    sentiment_numeric = [1 if label == "POSITIVE" else -1 for label in sentiment_scores]
-    axes[4].scatter(range(len(sentiment_numeric)), sentiment_numeric, c=sns.color_palette("coolwarm", as_cmap=True)(sentiment_numeric))
-    axes[4].set_title("Sentiment Trend (10 sec intervals)")
-    axes[4].set_xticks(range(0, len(sentiment_scores), len(sentiment_scores)//5))
-    axes[4].set_yticks([-1, 1], labels=["Negative", "Positive"])
-
-    # **Plot 6: Performance Scoring**
-    duration_min = librosa.get_duration(y=y, sr=sr) / 60
-    if duration_min > 8:
-        performance = "Critical (Long call, needs improvement)"
-        box_color = "#FFCCCB"  # Light red for critical calls
-    elif sentiment_label == "NEGATIVE":
-        performance = "Needs Improvement (Negative sentiment)"
-        box_color = "#FFDD00"  # Light yellow for improvement needed
-    else:
-        performance = "Good (Positive sentiment)"
-        box_color = "#98FB98"  # Light green for good performance
-
-    # Hide the axis
-    axes[5].axis('off')
-
-    # Add a colored rectangle as the background box for performance text
-    box = patches.FancyBboxPatch((0.05, 0.4), 0.9, 0.2, boxstyle="round,pad=0.05", linewidth=2, edgecolor="black", facecolor=box_color)
-    axes[5].add_patch(box)
-
-    # Add the performance text inside the colored box
-    axes[5].text(0.5, 0.5, f"Agent Performance: {performance}", ha='center', va='center', fontsize=14, color="black")
-
-    # Adjust the spacing between the subplots to avoid overlap
-    plt.subplots_adjust(wspace=0.4, hspace=0.5)
-
-    # Display the plot in Streamlit
+    sentiment_numeric = [1 if s == "POSITIVE" else -1 for s in sentiment_scores]
+
+    # Plot sentiment trend
+    fig, ax = plt.subplots(figsize=(8, 4))
+    ax.scatter(range(len(sentiment_numeric)), sentiment_numeric, c=sentiment_numeric, cmap="coolwarm")
+    ax.set_title("Sentiment Trend (10-sec intervals)")
+    ax.set_xticks(range(0, len(sentiment_scores), max(1, len(sentiment_scores)//5)))
+    ax.set_yticks([-1, 1], labels=["Negative", "Positive"])
     st.pyplot(fig)
 
-    # Display words from audio (positive and negative words in boxes)
+    # Display positive & negative words
     st.subheader("🗣️ Positive and Negative Words in Audio")
-
-    # Create two horizontal boxes for good and negative words
-    col1, col2 = st.columns([1, 1])
+    col1, col2 = st.columns(2)
     with col1:
         st.markdown("### Good Words")                           
         st.write(", ".join(good_words) if good_words else "No good words detected.")
@@ -179,13 +159,9 @@ if uploaded_file:
         st.markdown("### Negative Words")
         st.write(", ".join(negative_words) if negative_words else "No negative words detected.")
 
-    # Display only negative words separately
-    st.subheader("🗣️ Negative Words in Audio")
-    st.write(", ".join(negative_words) if negative_words else "No negative words detected.")
+    # Display full transcription
+    st.subheader("📝 Full Transcription")
+    st.write(transcribed_text)
 
-    # Display transcription result in Streamlit
-    st.subheader("📝 Full Transcription Result (Raw JSON)")
-    st.json(result)
-    
     # Clean up temp file
     os.remove(file_path)