pelican-nlp 0.1.1__py3-none-any.whl → 0.1.3__py3-none-any.whl

pelican_nlp/Nils_backup/transcription/annotation_tool_boundaries.py

@@ -0,0 +1,1122 @@
+import sys
+import os
+import json
+import tempfile
+import numpy as np
+import librosa
+import time
+
+from PyQt5.QtWidgets import (
+    QApplication, QMainWindow, QWidget, QTableWidget, QTableWidgetItem,
+    QVBoxLayout, QHBoxLayout, QPushButton, QComboBox, QFileDialog, QMessageBox,
+    QInputDialog, QMenu, QAction, QAbstractItemView, QSplitter, QUndoStack, QUndoCommand, QScrollBar
+)
+from PyQt5.QtCore import Qt, QTimer, QObject, pyqtSignal, QThread
+from PyQt5.QtGui import QColor
+
+import pyqtgraph as pg
+from pydub import AudioSegment
+from pydub.playback import _play_with_simpleaudio as play_audio
+
+# --- Undo/Redo Command Classes ---
+
+class AddRowCommand(QUndoCommand):
+    def __init__(self, main_window, row_position, row_data, description="Add Row"):
+        super().__init__(description)
+        self.main_window = main_window
+        self.row_position = row_position
+        self.row_data = row_data
+
+    def redo(self):
+        self.main_window.insert_row(self.row_position, self.row_data)
+
+    def undo(self):
+        self.main_window.remove_row(self.row_position)
+
+
+class DeleteRowsCommand(QUndoCommand):
+    def __init__(self, main_window, rows_data, row_positions, description="Delete Rows"):
+        super().__init__(description)
+        self.main_window = main_window
+        self.rows_data = rows_data
+        self.row_positions = row_positions
+
+    def redo(self):
+        for row in sorted(self.row_positions, reverse=True):
+            self.main_window.remove_row(row)
+
+    def undo(self):
+        for row, data in sorted(zip(self.row_positions, self.rows_data)):
+            self.main_window.insert_row(row, data)
+
+
+class EditCellCommand(QUndoCommand):
+    def __init__(self, main_window, row, column, old_value, new_value, description="Edit Cell"):
+        super().__init__(description)
+        self.main_window = main_window
+        self.row = row
+        self.column = column
+        self.old_value = old_value
+        self.new_value = new_value
+
+    def redo(self):
+        self.main_window.set_cell(self.row, self.column, self.new_value)
+
+    def undo(self):
+        self.main_window.set_cell(self.row, self.column, self.old_value)
+
+
+class BulkEditSpeakerCommand(QUndoCommand):
+    def __init__(self, main_window, row_positions, old_speakers, new_speaker, description="Bulk Edit Speaker"):
+        super().__init__(description)
+        self.main_window = main_window
+        self.row_positions = row_positions
+        self.old_speakers = old_speakers
+        self.new_speaker = new_speaker
+
+    def redo(self):
+        for row in self.row_positions:
+            self.main_window.set_speaker(row, self.new_speaker)
+
+    def undo(self):
+        for row, speaker in zip(self.row_positions, self.old_speakers):
+            self.main_window.set_speaker(row, speaker)
+
+
+# --- Audio Loader for Asynchronous Loading ---
+
+class AudioLoader(QObject):
+    finished = pyqtSignal(np.ndarray, int)
+    error = pyqtSignal(str)
+
+    def __init__(self, file_path, downsample_factor=100):
+        super().__init__()
+        self.file_path = file_path
+        self.downsample_factor = downsample_factor
+
+    def run(self):
+        try:
+            # Load audio using librosa for consistent sampling rate
+            y, sr = librosa.load(self.file_path, sr=None, mono=True)
+            duration = librosa.get_duration(y=y, sr=sr)
+            samples = y
+
+            # Normalize samples
+            max_abs_sample = np.max(np.abs(samples))
+            samples = samples / max_abs_sample if max_abs_sample != 0 else samples
+
+            # Downsample if necessary
+            if len(samples) > 1_000_000:
+                samples = self.downsample_waveform(samples, self.downsample_factor)
+
+            self.finished.emit(samples, sr)
+        except Exception as e:
+            self.error.emit(str(e))
+
+    def downsample_waveform(self, samples, factor):
+        num_blocks = len(samples) // factor
+        return np.array([samples[i * factor:(i + 1) * factor].mean() for i in range(num_blocks)])
+
+
+# --- Draggable Line Class for pyqtgraph ---
+
+class DraggableLine(pg.InfiniteLine):
+    def __init__(self, pos, color, idx, boundary_type, pen=None, movable=True):
+        pen = pen or pg.mkPen(color=color, width=2)
+        super().__init__(pos=pos, angle=90, pen=pen, movable=movable)
+        self.idx = idx
+        self.boundary_type = boundary_type
+        self.setHoverPen(pen.color().lighter())
+        self.setCursor(Qt.SizeHorCursor)
+
+
+# --- WaveformCanvas Class Using pyqtgraph ---
+
+class WaveformCanvas(QWidget):
+    boundary_changed = pyqtSignal(int, str, float)  # idx, 'start'/'end', new position
+    waveform_clicked = pyqtSignal(float)
+    audio_loaded = pyqtSignal()
+    loading_error = pyqtSignal(str)
+
+    def __init__(self, parent=None):
+        super().__init__(parent)
+        self.layout = QVBoxLayout(self)
+        self.plot_widget = pg.PlotWidget()
+        self.plot_widget.setYRange(-1, 1)
+        self.plot_widget.showGrid(x=True, y=False)
+        self.plot_widget.setLabel('bottom', 'Time', 's')
+        self.layout.addWidget(self.plot_widget)
+
+        self.scrollbar = QScrollBar(Qt.Horizontal)
+        self.layout.addWidget(self.scrollbar)
+        self.scrollbar.valueChanged.connect(self.on_scrollbar_value_changed)
+
+        self.plot_widget.plotItem.vb.sigXRangeChanged.connect(self.on_x_range_changed)
+
+        self.words = []
+        self.lines = []
+        self.connecting_lines = []
+
+        self.dragging_line = None
+        
+        self.utterances = []
+        self.utterance_items = []
+        self.utterance_regions = []
+
+        self.plot_widget.scene().sigMouseClicked.connect(self.on_waveform_click)
+
+        self.audio_data = None
+        self.sr = None
+        self.duration = None
+        self.window_size = 5.0
+
+    def load_audio(self, file_path):
+        self.thread = QThread()
+        self.loader = AudioLoader(file_path)
+        self.loader.moveToThread(self.thread)
+        self.thread.started.connect(self.loader.run)
+        self.loader.finished.connect(self.on_audio_loaded)
+        self.loader.finished.connect(self.thread.quit)
+        self.loader.finished.connect(self.loader.deleteLater)
+        self.thread.finished.connect(self.thread.deleteLater)
+        self.loader.error.connect(self.on_loading_error)
+        self.thread.start()
+
+    def on_audio_loaded(self, samples, sr):
+        self.audio_data = samples
+        self.sr = sr
+        effective_sr = sr /  self.loader.downsample_factor
+        self.duration = len(samples) / effective_sr
+        t = np.linspace(0, self.duration, num=len(samples))
+
+        self.plot_widget.clear()
+        self.plot_widget.plot(t, samples, pen='b')
+        self.playtime_line = pg.InfiniteLine(pos=0, angle=90, pen=pg.mkPen('y', width=4))
+        self.playtime_line.setZValue(1000)
+        self.plot_widget.addItem(self.playtime_line)
+        self.plot_widget.setLimits(xMin=0, xMax=self.duration)
+        self.plot_widget.setXRange(0, min(self.window_size, self.duration))
+        self.plot_widget.setLabel('bottom', 'Time', 's')
+        self.draw_lines()
+        self.boundary_changed.emit(-1, '', 0.0)  # Reset
+        print("Audio loaded successfully.")
+        print(f"Duration: {self.duration}, Window Size: {self.window_size}")
+        self.scrollbar.setMinimum(0)
+        self.scrollbar.setMaximum(int(self.duration * 1000))
+        self.scrollbar.setSingleStep(100)
+        self.scrollbar.setPageStep(int(self.window_size * 1000))
+        self.scrollbar.setValue(0)
+
+        self.audio_loaded.emit()
+        
+    def load_utterances(self, utterances):
+        self.utterances = utterances
+        self.draw_utterances()
+        
+    def draw_utterances(self):
+        self.clear_utterance_items()
+        self.utterance_items = []
+        self.utterance_regions = []
+
+        speaker_colors = {
+            "SPEAKER_00": QColor(255, 200, 200, 100),  # Light red
+            "SPEAKER_01": QColor(200, 255, 200, 100),  # Light green
+            "SPEAKER_02": QColor(200, 200, 255, 100),  # Light blue
+            "UNKNOWN": QColor(200, 200, 200, 100),
+            "": QColor(200, 200, 200, 100),  # Light gray
+        }
+
+        for idx, utterance in enumerate(self.utterances):
+            start = float(utterance['start_time'])
+            end = float(utterance['end_time'])
+            speaker = utterance.get('speaker', '')
+            confidence = utterance.get('confidence', '')
+            color = speaker_colors.get(speaker, QColor(200, 200, 200, 100))
+
+            # Add background region for utterance
+            region = pg.LinearRegionItem(values=[start+0.01, end-0.01], brush=color)
+            region.setMovable(False)
+            self.plot_widget.addItem(region)
+            self.utterance_regions.append(region)
+
+            # Break utterance text into words
+            text = utterance.get('text', '')
+            words = text.strip().split()
+
+            num_words = len(words)
+            if num_words == 0:
+                continue
+
+            # Add a label for utterance metadata (e.g., speaker, index, duration)
+            label_text = f"Utterance: {idx + 1}, Speaker: {speaker}, Speaker Confidence: {confidence}, Duration: {round(end - start, 2)}s"
+            meta_label = pg.TextItem(label_text, anchor=(0.5, 0), color='yellow')
+            meta_label.setPos((start + end) / 2, -0.25)  # Centered above the utterance
+            self.plot_widget.addItem(meta_label)
+            self.utterance_items.append(meta_label)
+
+            word_times = np.linspace(start, end, num_words+2)
+
+            for word, word_time in zip(words, word_times[1:-1]):
+                # Plot the word as a text label at y=-0.5
+                label = pg.TextItem(word, anchor=(0.5, 0), color='white')
+                label.setPos(word_time, -0.5)
+                self.plot_widget.addItem(label)
+                self.utterance_items.append(label)
+            
+    def clear_utterance_items(self):
+        if hasattr(self, 'utterance_items'):
+            for item in self.utterance_items:
+                self.plot_widget.removeItem(item)
+            self.utterance_items = []
+        if hasattr(self, 'utterance_regions'):
+            for region in self.utterance_regions:
+                self.plot_widget.removeItem(region)
+            self.utterance_regions = []
+
+    def on_loading_error(self, error_message):
+        self.loading_error.emit(error_message)
+
+    def load_words(self, words):
+        self.words = words
+        self.draw_lines()
+
+    def clear_lines(self):
+        for line in self.lines:
+            self.plot_widget.removeItem(line['line'])
+        self.lines = []
+        for cline in self.connecting_lines:
+            self.plot_widget.removeItem(cline)
+        self.connecting_lines = []
+
+    def draw_lines(self):
+        self.clear_lines()
+        for idx, word in enumerate(self.words):
+            
+            if idx % 2 == 0:
+                y_pos_line = 0.55
+            else:
+                y_pos_line = 0.45
+            
+            ### I adjust the line positions slightly because otherwise start and endlines of consecutive words ###
+            start_line = DraggableLine(pos=(word['start']+0.005), color='green', idx=idx, boundary_type='start')
+            end_line = DraggableLine(pos=(word['end']-0.005), color='red', idx=idx, boundary_type='end')
+            self.plot_widget.addItem(start_line)
+            self.plot_widget.addItem(end_line)
+            self.lines.append({'line': start_line, 'idx': idx, 'type': 'start'})
+            self.lines.append({'line': end_line, 'idx': idx, 'type': 'end'})
+
+            # Connecting line at y=0.5
+            connecting_line = pg.PlotCurveItem(
+                [word['start'] +0.005, word['end']-0.005],
+                [y_pos_line, y_pos_line],  # Position the line at y=0.5
+                pen=pg.mkPen('blue', width=2),
+            )
+            self.plot_widget.addItem(connecting_line)
+
+            # Create arrowheads
+            start_arrow = self.create_arrow(word['start'] + 0.005, y_pos_line, 0)
+            end_arrow = self.create_arrow(word['end'] - 0.005, y_pos_line, 180)
+
+            # Create label
+            label = pg.TextItem(word['word'], anchor=(0.5, 0), color='white')
+            self.plot_widget.addItem(label)
+
+            # Store all items in the connecting_lines list
+            self.connecting_lines.append({
+                "line": connecting_line,
+                "start_arrow": start_arrow,
+                "end_arrow": end_arrow,
+                "label": label,
+            })
+
+            # Position label initially
+            self.update_connecting_line(idx)
+
+            # Connect signals to update arrows and labels
+            start_line.sigPositionChangeFinished.connect(lambda _, line=start_line: self.on_line_moved(line))
+            end_line.sigPositionChangeFinished.connect(lambda _, line=end_line: self.on_line_moved(line))
+
+        self.plot_widget.update()
+    
+    def create_arrow(self, x, y, angle):
+        
+        arrow = pg.ArrowItem(
+            pos=(x, y),
+            angle=angle,  # Direction of the arrow in degrees
+            tipAngle=30,
+            baseAngle=20,
+            headLen=15,
+            brush='blue',
+        )
+        self.plot_widget.addItem(arrow)
+        return arrow
+
+    def update_connecting_line(self, idx):
+        word = self.words[idx]
+        start = word['start']
+        end = word['end']
+        
+        if idx % 2 == 0:
+            y_pos_line = 0.55
+        else:
+            y_pos_line = 0.45
+
+        # Update the connecting line's x-coordinates and keep y fixed at 0.5
+        self.connecting_lines[idx]['line'].setData([start+ 0.005, end- 0.005], [y_pos_line, y_pos_line])
+
+        # Update arrowhead positions
+        self.connecting_lines[idx]['start_arrow'].setPos(start+ 0.005, y_pos_line)
+        self.connecting_lines[idx]['end_arrow'].setPos(end- 0.005, y_pos_line)
+
+        # Update label position (middle of the line, slightly above)
+        mid_x = (start + end) / 2
+        
+        if word["speaker"] in ["", "UNKOWN"]:
+            mid_y = 0.7  # Slightly above y=0.5
+        elif word["speaker"] == "SPEAKER_00":
+            mid_y = 0.4
+        elif word["speaker"] == "SPEAKER_01":
+            mid_y = 0.6
+        else:
+            mid_y = 0.0
+        
+        self.connecting_lines[idx]['label'].setPos(mid_x, mid_y)
+
+    def on_line_moved(self, line):
+        idx = line.idx
+        boundary_type = line.boundary_type
+        new_pos = line.value()
+        new_pos = max(0.0, min(new_pos, self.duration))
+        self.words[idx][boundary_type] = new_pos
+        self.boundary_changed.emit(idx, boundary_type, new_pos)
+        self.update_connecting_line(idx)
+
+    def on_waveform_click(self, event):
+        pos = event.scenePos()
+        if not self.plot_widget.sceneBoundingRect().contains(pos):
+            return
+        mouse_point = self.plot_widget.plotItem.vb.mapSceneToView(pos)
+        clicked_time = mouse_point.x()
+        clicked_time = max(0.0, min(clicked_time, self.duration))
+        self.waveform_clicked.emit(clicked_time)
+
+    def update_playtime_line(self, current_time):
+        self.playtime_line.setPos(current_time)
+        # Adjust view range
+        self.adjust_view_range(current_time)
+
+    def add_word_lines(self, row, start_time, end_time):
+        """Add draggable lines for a new word."""
+        start_line = DraggableLine(pos=start_time, color='green', idx=row, boundary_type='start')
+        end_line = DraggableLine(pos=end_time, color='red', idx=row, boundary_type='end')
+        self.plot_widget.addItem(start_line)
+        self.plot_widget.addItem(end_line)
+        self.lines.append({'line': start_line, 'idx': row, 'type': 'start'})
+        self.lines.append({'line': end_line, 'idx': row, 'type': 'end'})
+
+        # Connecting line with arrows
+        connecting_line = pg.ArrowItem(
+            pos=((start_time + end_time) / 2, 0),
+            angle=0,
+            tipAngle=30,
+            baseAngle=20,
+            headLen=15,
+            tailLen=0,
+            tailWidth=0,
+            brush='blue'
+        )
+        connecting_line.setParentItem(self.plot_widget.plotItem)
+        self.connecting_lines.append(connecting_line)
+
+        start_line.sigPositionChangeFinished.connect(lambda _, line=start_line: self.on_line_moved(line))
+        end_line.sigPositionChangeFinished.connect(lambda _, line=end_line: self.on_line_moved(line))
+
+    def adjust_view_range(self, current_time, window_size=None):
+        if window_size is None:
+            window_size = self.window_size
+        half_window = window_size / 2.0
+        start = max(0.0, current_time - half_window)
+        end = min(self.duration, current_time + half_window)
+        self.plot_widget.setXRange(start, end, padding=0)
+        self.scrollbar.blockSignals(True)
+        self.scrollbar.setValue(int(start * 1000))
+        self.scrollbar.blockSignals(False)
+
+    def update_line_position(self, idx, boundary_type, new_pos):
+        # Find the line and update its position
+        for line_info in self.lines:
+            if line_info['idx'] == idx and line_info['type'] == boundary_type:
+                line_info['line'].setValue(new_pos)
+                break
+        self.words[idx][boundary_type] = new_pos
+        self.update_connecting_line(idx)
+
+    def on_scrollbar_value_changed(self, value):
+        start = min(value / 1000.0, self.duration - self.window_size)
+        end = min(start + self.window_size, self.duration)
+        self.plot_widget.setXRange(start, end, padding=0)
+
+    def on_x_range_changed(self, view_box, range):
+        start, end = max(0, range[0]), min(self.duration, range[1])
+        self.scrollbar.blockSignals(True)
+        self.scrollbar.setValue(int(start * 1000))
+        self.scrollbar.blockSignals(False)
+
+
+# --- MainWindow Class ---
+
+class MainWindow(QMainWindow):
+    def __init__(self):
+        super().__init__()
+        self.setWindowTitle("PELICAn Transcription Tool")
+        self.setGeometry(100, 100, 1600, 900)
+
+        # Initialize variables
+        self.audio_segment = None
+        self.is_playing = False
+        self.play_obj = None
+        self.current_time = 0.0
+        self.speakers = []
+        self.undo_stack = QUndoStack(self)
+        self.old_values = {}
+        self.temp_file_path = tempfile.NamedTemporaryFile(delete=False, suffix='.json').name
+        self.previous_current_row = None  # Initialize previous_current_row
+        # Setup UI components
+        self.setup_ui()
+        self.setup_signals()
+        
+        # Start the autosave timer
+        self.autosave_timer = QTimer(self)
+        self.autosave_timer.timeout.connect(self.autosave)
+        self.autosave_timer.start(5000)  # Trigger autosave every 5 seconds
+
+        # Load autosave if exists
+        self.load_autosave()
+
+    def setup_ui(self):
+        splitter = QSplitter(Qt.Horizontal)
+
+        # Left panel: Waveform
+        self.waveform_widget = QWidget()
+        waveform_layout = QVBoxLayout(self.waveform_widget)
+        self.canvas = WaveformCanvas(parent=self.waveform_widget)
+        waveform_layout.addWidget(self.canvas)
+
+        # Playback Controls
+        playback_layout = QHBoxLayout()
+        self.play_button = QPushButton("Play")
+        self.play_button.clicked.connect(self.toggle_playback)
+        self.stop_button = QPushButton("Stop")
+        self.stop_button.clicked.connect(self.stop_playback)
+        self.return_button = QPushButton("Return to Selection (X)")
+        self.return_button.clicked.connect(self.return_to_selection)
+        playback_layout.addWidget(self.play_button)
+        playback_layout.addWidget(self.stop_button)
+        playback_layout.addWidget(self.return_button)
+        waveform_layout.addLayout(playback_layout)
+
+        # Load and Save Buttons
+        buttons_layout = QHBoxLayout()
+        load_audio_button = QPushButton("Load Audio")
+        load_audio_button.clicked.connect(self.load_audio)
+        load_transcript_button = QPushButton("Load Transcript")
+        load_transcript_button.clicked.connect(self.load_transcript)
+        save_button = QPushButton("Save Annotations")
+        save_button.clicked.connect(self.save_annotations)
+        buttons_layout.addWidget(load_audio_button)
+        buttons_layout.addWidget(load_transcript_button)
+        buttons_layout.addWidget(save_button)
+        waveform_layout.addLayout(buttons_layout)
+
+        # Undo/Redo Buttons
+        undo_redo_layout = QHBoxLayout()
+        undo_button = QPushButton("Undo")
+        undo_button.clicked.connect(self.undo_stack.undo)
+        redo_button = QPushButton("Redo")
+        redo_button.clicked.connect(self.undo_stack.redo)
+        undo_redo_layout.addWidget(undo_button)
+        undo_redo_layout.addWidget(redo_button)
+        waveform_layout.addLayout(undo_redo_layout)
+
+        splitter.addWidget(self.waveform_widget)
+
+        # Right panel: Transcript Table
+        self.table_widget = QTableWidget()
+        self.table_widget.setColumnCount(4)
+        self.table_widget.setHorizontalHeaderLabels(["Word", "Start Time", "End Time", "Speaker"])
+        self.table_widget.setSelectionBehavior(QAbstractItemView.SelectRows)
+        self.table_widget.setSelectionMode(QAbstractItemView.ExtendedSelection)
+        self.table_widget.setContextMenuPolicy(Qt.CustomContextMenu)
+        self.table_widget.customContextMenuRequested.connect(self.show_context_menu)
+        splitter.addWidget(self.table_widget)
+
+        self.setCentralWidget(splitter)
+
+    def setup_signals(self):
+        self.canvas.boundary_changed.connect(self.on_boundary_changed)
+        self.canvas.waveform_clicked.connect(self.on_waveform_clicked)
+        self.canvas.audio_loaded.connect(self.on_audio_loaded)
+        self.canvas.loading_error.connect(self.on_audio_load_error)
+        self.table_widget.itemChanged.connect(self.on_item_changed)
+        self.table_widget.cellDoubleClicked.connect(self.on_cell_double_clicked)
+        self.table_widget.currentCellChanged.connect(self.on_current_cell_changed)
+        self.table_widget.selectionModel().selectionChanged.connect(self.on_selection_changed)
+
+    def load_audio(self):
+        options = QFileDialog.Options()
+        file_path, _ = QFileDialog.getOpenFileName(
+            self,
+            "Open Audio File",
+            "",
+            "Audio Files (*.wav *.mp3 *.flac *.ogg);;All Files (*)",
+            options=options,
+        )
+        if file_path:
+            self.canvas.load_audio(file_path)
+            # Load audio segment for playback
+            try:
+                self.audio_segment = AudioSegment.from_file(file_path).set_channels(1)
+            except Exception as e:
+                QMessageBox.critical(self, "Audio Load Error", f"Failed to load audio for playback:\n{str(e)}")
+
+    def on_audio_loaded(self):
+        self.statusBar().showMessage("Audio loaded successfully.", 5000)
+        self.canvas.adjust_view_range(self.current_time)
+
+    def on_audio_load_error(self, error_message):
+        QMessageBox.critical(self, "Audio Load Error", f"Failed to load audio file:\n{error_message}")
+        self.statusBar().showMessage("Failed to load audio.", 5000)
+
+    def load_transcript(self):
+        options = QFileDialog.Options()
+        file_path, _ = QFileDialog.getOpenFileName(
+            self,
+            "Open Transcript File",
+            "",
+            "JSON Files (*.json);;All Files (*)",
+            options=options,
+        )
+        if file_path:
+            try:
+                with open(file_path, 'r', encoding='utf-8') as f:
+                    transcript = json.load(f)
+                    words = transcript.get("combined_data", [])
+                    speaker_segments = transcript.get("speaker_segments", [])
+                    utterances = transcript.get("utterance_data", [])
+                    self.canvas.load_utterances(utterances)
+                # Validate and process words
+                for word in words:
+                    if 'word' not in word or 'start_time' not in word or 'end_time' not in word:
+                        raise ValueError("Invalid transcript format. Each entry must have 'word', 'start_time', and 'end_time'.")
+                    word['start'] = float(word['start_time'])
+                    word['end'] = float(word['end_time'])
+                    if 'speaker' not in word:
+                        word['speaker'] = ''
+                    else:
+                        speaker = word['speaker']
+                        if speaker and speaker not in self.speakers:
+                            self.speakers.append(speaker)
+                self.canvas.load_words(words)
+                self.update_table()
+                print(f"Loaded transcript file: {file_path}")
+            except Exception as e:
+                QMessageBox.critical(self, "Error Loading Transcript", f"Failed to load transcript:\n{str(e)}")
+
+    def update_table(self):
+        self.table_widget.blockSignals(True)
+        words = self.canvas.words
+        self.table_widget.setRowCount(len(words))
+        for i, word in enumerate(words):
+            word_item = QTableWidgetItem(word['word'])
+            start_item = QTableWidgetItem(f"{word['start']:.2f}")
+            end_item = QTableWidgetItem(f"{word['end']:.2f}")
+            speaker_dropdown = QComboBox()
+            speaker_dropdown.addItems(self.speakers + [""])
+            speaker_dropdown.setCurrentText(word.get('speaker', ''))
+            speaker_dropdown.currentTextChanged.connect(self.on_speaker_changed)
+            self.table_widget.setItem(i, 0, word_item)
+            self.table_widget.setItem(i, 1, start_item)
+            self.table_widget.setItem(i, 2, end_item)
+            self.table_widget.setCellWidget(i, 3, speaker_dropdown)
+            # Set default cell colors
+            for j in range(3):
+                item = self.table_widget.item(i, j)
+                if item:
+                    item.setBackground(QColor("black"))
+                    item.setForeground(QColor("white"))
+        self.table_widget.blockSignals(False)
+        self.statusBar().showMessage("Transcript loaded successfully.", 5000)
+
+    def on_boundary_changed(self, idx, boundary_type, new_pos):
+        if idx == -1:
+            return
+        self.table_widget.blockSignals(True)
+        if boundary_type == 'start':
+            item = self.table_widget.item(idx, 1)
+            if item is not None:
+                item.setText(f"{new_pos:.2f}")
+        elif boundary_type == 'end':
+            item = self.table_widget.item(idx, 2)
+            if item is not None:
+                item.setText(f"{new_pos:.2f}")
+        self.table_widget.blockSignals(False)
+        self.autosave()
+
+    def on_waveform_clicked(self, time):
+        self.current_time = time
+        self.canvas.update_playtime_line(self.current_time)
+        self.highlight_current_row()
+
+    def on_item_changed(self, item):
+        if self.table_widget.signalsBlocked():
+            return
+        row = item.row()
+        column = item.column()
+        key = (row, column)
+        old_value = self.old_values.get(key, "")
+        new_value = item.text()
+
+        if old_value != new_value:
+            if column in [1, 2]:
+                try:
+                    new_time = float(new_value)
+                    boundary_type = 'start' if column == 1 else 'end'
+                    self.canvas.update_line_position(row, boundary_type, new_time)
+                except ValueError:
+                    QMessageBox.warning(self, "Invalid Input", "Start and End times must be numeric.")
+                    self.table_widget.blockSignals(True)
+                    item.setText(old_value)
+                    self.table_widget.blockSignals(False)
+                    return
+
+            command = EditCellCommand(self, row, column, old_value, new_value)
+            self.undo_stack.push(command)
+            self.autosave()
+
+        if key in self.old_values:
+            del self.old_values[key]
+
+    def on_cell_double_clicked(self, row, column):
+        item = self.table_widget.item(row, column)
+        if item:
+            self.old_values[(row, column)] = item.text()
+
+    def on_speaker_changed(self, new_speaker):
+        sender = self.sender()
+        if new_speaker and new_speaker not in self.speakers:
+            self.speakers.append(new_speaker)
+            self.update_speaker_dropdowns()
+
+    def update_speaker_dropdowns(self):
+        for row in range(self.table_widget.rowCount()):
+            speaker_dropdown = self.table_widget.cellWidget(row, 3)
+            if speaker_dropdown:
+                current_speaker = speaker_dropdown.currentText()
+                speaker_dropdown.blockSignals(True)
+                speaker_dropdown.clear()
+                speaker_dropdown.addItems(self.speakers + [""])
+                speaker_dropdown.setCurrentText(current_speaker)
+                speaker_dropdown.blockSignals(False)
+
+    def toggle_playback(self):
+        if self.audio_segment is None:
+            QMessageBox.warning(self, "No Audio", "Please load an audio file first.")
+            return
+        if self.is_playing:
+            self.pause_playback()
+        else:
+            self.start_playback()
+
+    def start_playback(self):
+        if self.audio_segment is not None:
+            try:
+                start_ms = int(self.current_time * 1000)
+                sliced_audio = self.audio_segment[start_ms:]
+                self.play_obj = play_audio(sliced_audio)
+                self.is_playing = True
+                self.play_button.setText("Pause")
+                # Record the start time and position
+                self.playback_start_time = time.time()
+                self.playback_start_position = self.current_time
+                self.playback_timer = QTimer()
+                self.playback_timer.timeout.connect(self.update_current_time)
+                self.playback_timer.start(50)  # Increased frequency for smoother updates
+            except Exception as e:
+                QMessageBox.critical(self, "Playback Error", f"Failed to play audio:\n{str(e)}")
+
+    def pause_playback(self):
+        if self.play_obj:
+            self.play_obj.stop()
+            self.play_obj = None
+        self.is_playing = False
+        self.play_button.setText("Play")
+        if hasattr(self, 'playback_timer') and self.playback_timer.isActive():
+            self.playback_timer.stop()
+        # Update current_time to the actual playback position
+        elapsed_time = time.time() - self.playback_start_time
+        self.current_time = self.playback_start_position + elapsed_time
+
+    def stop_playback(self):
+        if self.play_obj:
+            self.play_obj.stop()
+            self.play_obj = None
+        self.is_playing = False
+        self.play_button.setText("Play")
+        self.current_time = 0.0
+        self.canvas.update_playtime_line(self.current_time)
+        self.highlight_current_row()
+        if hasattr(self, 'playback_timer') and self.playback_timer.isActive():
+            self.playback_timer.stop()
+
+    def update_current_time(self):
+        elapsed_time = time.time() - self.playback_start_time
+        self.current_time = self.playback_start_position + elapsed_time
+        if self.current_time >= self.canvas.duration:
+            self.stop_playback()
+            return
+        self.canvas.update_playtime_line(self.current_time)
+        self.highlight_current_row()
+
+    def highlight_current_row(self):
+        current_row = self.get_current_row()
+        selected_rows = self.get_selected_rows()
+        
+        if current_row == self.previous_current_row:
+            return  # No change, so no need to update
+        
+        # Reset previous current row background
+        if self.previous_current_row is not None:
+            for column in range(3):
+                item = self.table_widget.item(self.previous_current_row, column)
+                if item:
+                    if self.previous_current_row in selected_rows:
+                        item.setBackground(QColor("blue"))
+                        item.setForeground(QColor("white"))
+                    else:
+                        item.setBackground(QColor("black"))
+                        item.setForeground(QColor("white"))
+        
+        # Set new current row background
+        if current_row != -1:
+            for column in range(3):
+                item = self.table_widget.item(current_row, column)
+                if item:
+                    item.setBackground(QColor("yellow"))
+                    item.setForeground(QColor("black"))
+            self.table_widget.scrollToItem(self.table_widget.item(current_row, 0), QAbstractItemView.PositionAtCenter)
+        
+        self.previous_current_row = current_row
+    
+    
+    def get_current_row(self):
+        for row in range(self.table_widget.rowCount()):
+            try:
+                start_time = float(self.table_widget.item(row, 1).text())
+                end_time = float(self.table_widget.item(row, 2).text())
+                if start_time <= self.current_time < end_time:
+                    return row
+            except (ValueError, AttributeError):
+                continue
+        return -1
+
+    def return_to_selection(self):
+        selected_rows = self.get_selected_rows()
+        if selected_rows:
+            first_row = selected_rows[0]
+            try:
+                start_time = float(self.table_widget.item(first_row, 1).text())
+                self.current_time = start_time
+                self.canvas.update_playtime_line(self.current_time)
+                self.highlight_current_row()
+            except (ValueError, AttributeError):
+                pass
+
+    def get_selected_rows(self):
+        return sorted(set(index.row() for index in self.table_widget.selectionModel().selectedRows()))
+
+    def show_context_menu(self, position):
+        menu = QMenu()
+        add_row_action = QAction("Add Row", self)
+        add_row_action.triggered.connect(self.add_row)
+        delete_row_action = QAction("Delete Selected Rows", self)
+        delete_row_action.triggered.connect(self.delete_selected_rows)
+        bulk_edit_action = QAction("Bulk Edit Speaker", self)
+        bulk_edit_action.triggered.connect(self.bulk_edit_speaker)
+        menu.addAction(add_row_action)
+        menu.addAction(delete_row_action)
+        menu.addAction(bulk_edit_action)
+        menu.exec_(self.table_widget.viewport().mapToGlobal(position))
+
+    def add_row(self):
+        row_count = self.table_widget.rowCount()
+        row_data = {'word': '', 'start_time': 0.0, 'end_time': 0.0, 'speaker': ''}
+        command = AddRowCommand(self, row_count, row_data)
+        self.undo_stack.push(command)
+        self.autosave()
+
+    def delete_selected_rows(self):
+        selected_rows = self.get_selected_rows()
+        if not selected_rows:
+            QMessageBox.information(self, "No Selection", "Please select at least one row to delete.")
+            return
+
+        # Gather data
+        rows_data = []
+        for row in selected_rows:
+            row_data = {
+                'word': self.table_widget.item(row, 0).text(),
+                'start_time': self.table_widget.item(row, 1).text(),
+                'end_time': self.table_widget.item(row, 2).text(),
+                'speaker': self.table_widget.cellWidget(row, 3).currentText()
+            }
+            rows_data.append(row_data)
+
+        confirm = QMessageBox.question(
+            self,
+            "Confirm Deletion",
+            f"Are you sure you want to delete {len(selected_rows)} selected row(s)?",
+            QMessageBox.Yes | QMessageBox.No
+        )
+        if confirm == QMessageBox.Yes:
+            command = DeleteRowsCommand(self, rows_data, selected_rows)
+            self.undo_stack.push(command)
+            self.autosave()
+
+    def bulk_edit_speaker(self):
+        selected_rows = self.get_selected_rows()
+        if not selected_rows:
+            QMessageBox.information(self, "No Selection", "Please select at least one row to edit.")
+            return
+
+        speaker, ok = QInputDialog.getItem(
+            self,
+            "Select Speaker",
+            "Choose a speaker to assign to selected rows:",
+            self.speakers + [""],
+            0,
+            False
+        )
+        if ok:
+            old_speakers = [self.table_widget.cellWidget(row, 3).currentText() for row in selected_rows]
+            command = BulkEditSpeakerCommand(self, selected_rows, old_speakers, speaker)
+            self.undo_stack.push(command)
+            if speaker and speaker not in self.speakers:
+                self.speakers.append(speaker)
+                self.update_speaker_dropdowns()
+            self.autosave()
+
+    def autosave(self):
+        words = self.canvas.words
+        data_to_save = []
+        for word in words:
+            data_to_save.append({
+                'word': word['word'],
+                'start_time': word['start'],
+                'end_time': word['end'],
+                'speaker': word.get('speaker', '')
+            })
+        try:
+            with open(self.temp_file_path, 'w', encoding='utf-8') as f:
+                json.dump(data_to_save, f, indent=4)
+            print(f"Autosaved annotations to {self.temp_file_path}")
+        except Exception as e:
+            print(f"Autosave failed: {e}")
+
+    def load_autosave(self):
+        if os.path.exists(self.temp_file_path):
+            try:
+                with open(self.temp_file_path, "r", encoding='utf-8') as file:
+                    annotations = json.load(file)
+                self.canvas.load_words(annotations)
+                self.update_table()
+                QMessageBox.information(self, "Recovery", "Recovered annotations from autosave.")
+            except Exception as e:
+                print(f"Failed to recover autosave: {e}")
+
+    def closeEvent(self, event):
+        reply = QMessageBox.question(
+            self, 'Exit',
+            "Do you want to save your annotations before exiting?",
+            QMessageBox.Yes | QMessageBox.No | QMessageBox.Cancel,
+            QMessageBox.Yes
+        )
+
+        if reply == QMessageBox.Yes:
+            if not self.validate_annotations():
+                event.ignore()
+                return
+            self.save_annotations()
+            if os.path.exists(self.temp_file_path):
+                os.remove(self.temp_file_path)
+            event.accept()
+        elif reply == QMessageBox.No:
+            if os.path.exists(self.temp_file_path):
+                os.remove(self.temp_file_path)
+            event.accept()
+        else:
+            event.ignore()
+
+    def save_annotations(self):
+        if not self.validate_annotations():
+            return
+
+        annotations = []
+        for row_idx in range(self.table_widget.rowCount()):
+            try:
+                start = float(self.table_widget.item(row_idx, 1).text())
+                end = float(self.table_widget.item(row_idx, 2).text())
+                start = round(start, 2)
+                end = round(end, 2)
+            except (ValueError, AttributeError):
+                QMessageBox.warning(self, "Invalid Input", f"Invalid start or end time at row {row_idx + 1}.")
+                return
+
+            word = self.table_widget.item(row_idx, 0).text()
+            speaker = self.table_widget.cellWidget(row_idx, 3).currentText()
+            annotations.append({
+                "word": word,
+                "start_time": start,
+                "end_time": end,
+                "speaker": speaker if speaker else None
+            })
+
+        options = QFileDialog.Options()
+        file_path, _ = QFileDialog.getSaveFileName(
+            self,
+            "Save Annotations",
+            "",
+            "JSON Files (*.json);;All Files (*)",
+            options=options,
+        )
+        if file_path:
+            try:
+                with open(file_path, 'w', encoding='utf-8') as f:
+                    json.dump(annotations, f, indent=4)
+                QMessageBox.information(self, "Save Successful", f"Annotations saved to {file_path}")
+            except Exception as e:
+                QMessageBox.critical(self, "Save Error", f"Failed to save annotations:\n{str(e)}")
+
+    def on_current_cell_changed(self, current_row, current_column, previous_row, previous_column):
+        if current_row >= 0:
+            try:
+                start_time = float(self.table_widget.item(current_row, 1).text())
+                self.current_time = start_time
+                self.canvas.update_playtime_line(self.current_time)
+                self.highlight_current_row()
+            except (ValueError, AttributeError):
+                pass
+
+    def on_selection_changed(self, selected, deselected):
+        self.previous_current_row = None
+        self.highlight_current_row()
+
+    def keyPressEvent(self, event):
+        if event.key() == Qt.Key_X:
+            self.return_to_selection()
+        else:
+            super().keyPressEvent(event)
+
+    def insert_row(self, row_position, row_data):
+        self.table_widget.blockSignals(True)
+        self.table_widget.insertRow(row_position)
+        word_item = QTableWidgetItem(row_data['word'])
+        start_item = QTableWidgetItem(f"{float(row_data['start_time']):.2f}")
+        end_item = QTableWidgetItem(f"{float(row_data['end_time']):.2f}")
+        speaker_dropdown = QComboBox()
+        speaker_dropdown.addItems(self.speakers + [""])
+        speaker_dropdown.setCurrentText(row_data.get('speaker', ''))
+        speaker_dropdown.currentTextChanged.connect(self.on_speaker_changed)
+        self.table_widget.setItem(row_position, 0, word_item)
+        self.table_widget.setItem(row_position, 1, start_item)
+        self.table_widget.setItem(row_position, 2, end_item)
+        self.table_widget.setCellWidget(row_position, 3, speaker_dropdown)
+        # Set default cell colors
+        for j in range(3):
+            item = self.table_widget.item(row_position, j)
+            if item:
+                item.setBackground(QColor("black"))
+                item.setForeground(QColor("white"))
+        self.table_widget.blockSignals(False)
+        # Update waveform
+        self.canvas.words.insert(row_position, {
+            'word': row_data['word'],
+            'start': float(row_data['start_time']),
+            'end': float(row_data['end_time']),
+            'speaker': row_data.get('speaker', '')
+        })
+        self.canvas.draw_lines()
+
+    def remove_row(self, row_position):
+        self.table_widget.blockSignals(True)
+        self.table_widget.removeRow(row_position)
+        self.table_widget.blockSignals(False)
+        # Update waveform
+        del self.canvas.words[row_position]
+        self.canvas.draw_lines()
+
+    def set_cell(self, row, column, value):
+        self.table_widget.blockSignals(True)
+        item = self.table_widget.item(row, column)
+        if item:
+            item.setText(value)
+        self.table_widget.blockSignals(False)
+        if column in [1, 2]:
+            try:
+                new_time = float(value)
+                boundary_type = 'start' if column == 1 else 'end'
+                self.canvas.update_line_position(row, boundary_type, new_time)
+                self.autosave()
+            except ValueError:
+                pass
+
+    def set_speaker(self, row, speaker):
+        self.table_widget.blockSignals(True)
+        speaker_dropdown = self.table_widget.cellWidget(row, 3)
+        if speaker_dropdown:
+            speaker_dropdown.setCurrentText(speaker)
+        self.table_widget.blockSignals(False)
+        if speaker and speaker not in self.speakers:
+            self.speakers.append(speaker)
+            self.update_speaker_dropdowns()
+            self.autosave()
+
+    def validate_annotations(self):
+        sorted_rows = sorted(range(self.table_widget.rowCount()), key=lambda r: float(self.table_widget.item(r, 1).text()) if self.table_widget.item(r, 1).text() else 0.0)
+        for i in range(len(sorted_rows)):
+            row = sorted_rows[i]
+            start_item = self.table_widget.item(row, 1)
+            end_item = self.table_widget.item(row, 2)
+            if not start_item or not end_item:
+                QMessageBox.warning(self, "Invalid Annotation", f"Missing start or end time at row {row + 1}.")
+                return False
+            try:
+                start_time = float(start_item.text())
+                end_time = float(end_item.text())
+                if start_time > end_time:
+                    QMessageBox.warning(self, "Invalid Annotation", f"Start time must be less than end time at row {row + 1}.")
+                    return False
+                if i < len(sorted_rows) - 1:
+                    next_row = sorted_rows[i + 1]
+                    next_start = float(self.table_widget.item(next_row, 1).text())
+                    if end_time > next_start:
+                        QMessageBox.warning(
+                            self, "Invalid Annotation",
+                            f"Annotations at rows {row + 1} and {next_row + 1} overlap."
+                        )
+                        return False
+            except ValueError:
+                QMessageBox.warning(self, "Invalid Annotation", f"Non-numeric start or end time at row {row + 1}.")
+                return False
+        return True
+
+# --- Main Execution ---
+
+def main():
+    app = QApplication(sys.argv)
+    window = MainWindow()
+    window.show()
+    sys.exit(app.exec_())
+
+
+if __name__ == "__main__":
+    main()