# Filler Word Processing

Process filler word annotations to generate video edit lists.

- Repository: https://github.com/benchflow-ai/skillsbench (1888 stars)
- Category: Media
- License: Apache-2.0
- Last updated: 2026-10-24
- Runs scripts: no, instructions only

## Install

Claude Code: `npx skills add benchflow-ai/skillsbench --skill filler-word-processing -g -a claude-code -y`
Codex: `npx skills add benchflow-ai/skillsbench --skill filler-word-processing -g -a codex -y`
Cursor: `npx skills add benchflow-ai/skillsbench --skill filler-word-processing -g -a cursor -y`
Gemini CLI: `npx skills add benchflow-ai/skillsbench --skill filler-word-processing -g -a gemini-cli -y`

## Description

Process filler word annotations to generate video edit lists. Use when working with timestamp annotations for removing speech disfluencies (um, uh, like, you know) from audio/video content.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Filler Word Processing

## Annotation Format

Typical annotation JSON structure:
```json
[
  {"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:

```python
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
```

## Merging Overlapping Segments

When filler words are close together, merge their cut segments:

```python
def merge_overlapping_segments(segments, min_gap=0.1):
    """
    Merge segments that overlap or are very close together.

The rest: https://github.com/benchflow-ai/skillsbench/blob/HEAD/tasks-extra/video-filler-word-remover/environment/skills/filler-word-processing/SKILL.md