Filler Word Processing

Annotation Format

Typical annotation JSON structure:

[
  {"word": "um", "timestamp": 12.5},
  {"word": "like", "timestamp": 25.3},
  {"word": "you know", "timestamp": 45.8}
]

Converting Annotations to Cut Segments

Each filler word annotation marks when the word starts. To remove it, use word-specific durations since different fillers have different lengths:

import json

# Word-specific durations (in seconds)
WORD_DURATIONS = {
    "uh": 0.3,
    "um": 0.4,
    "hum": 0.6,
    "hmm": 0.6,
    "mhm": 0.55,
    "like": 0.3,
    "yeah": 0.35,
    "so": 0.25,
    "well": 0.35,
    "okay": 0.4,
    "basically": 0.55,
    "you know": 0.55,
    "i mean": 0.5,
    "kind of": 0.5,
    "i guess": 0.5,
}
DEFAULT_DURATION = 0.4

def annotations_to_segments(annotations_file, buffer=0.05):
    """
    Convert filler word annotations to (start, end) cut segments.

    Args:
        annotations_file: Path to JSON annotations
        buffer: Small buffer before the word (seconds)

    Returns:
        List of (start, end) tuples representing segments to remove
    """
    with open(annotations_file) as f:
        annotations = json.load(f)

    segments = []
    for ann in annotations:
        word = ann.get('word', '').lower().strip()
        timestamp = ann['timestamp']
        # Use word-specific duration, fall back to default
        word_duration = WORD_DURATIONS.get(word, DEFAULT_DURATION)
        # Cut starts slightly before the word
        start = max(0, timestamp - buffer)
        # Cut ends after word duration
        end = timestamp + word_duration
        segments.append((start, end))

    return segments