video-to-text

name	video-to-text
description	Tool for extracting speech text from video with timestamps. Use when asked to transcribe video, extract subtitles, generate transcript from video, or convert video speech to text.

Video to Text

Extracts speech text from video files using OpenAI Whisper and outputs a transcript with timestamps.

When to Use This Skill

• You need to extract spoken words from a video file
• You want a timestamped transcript for subtitles or analysis
• You need to convert video speech to text for further processing
• You have a video file and want to generate a text transcript

Prerequisites

• Python 3.8+
• ffmpeg installed and available in PATH
• Internet connection (for downloading Whisper models on first run)

Usage

The tool is located at tools/video-to-text/video-to-text.py.

Command Line Arguments

python video-to-text.py <VIDEO_PATH> <OUTPUT_DIR>

• VIDEO_PATH: Path to the input video file (supports most formats)
• OUTPUT_DIR: Directory where the transcript text file will be saved

Example

python tools/video-to-text/video-to-text.py "path/to/video.mp4" "output/folder"

The output file will be named after the video file (e.g., video.txt) and placed in the output directory.

Features

• Automatically creates output directory if it doesn't exist
• Extracts audio from video using ffmpeg (mono, 16kHz WAV)
• Transcribes using OpenAI Whisper (base model by default)
• Outputs timestamps in HH:mm:ss format
• Handles various video formats supported by ffmpeg
• Clean error messages and progress indicators

Output Format

Each line in the output text file has the format:

[HH:mm:ss] Transcribed text here

Example:

[00:00:00] Hello world
[00:00:02] This is a test

Troubleshooting

Unicode Path Issues on Windows

If the video file path contains non-ASCII characters (e.g., Japanese, Chinese), you may encounter encoding issues on Windows command prompt. Consider:

1. Renaming the file to use ASCII characters
2. Using short paths (8.3 format)
3. Running from PowerShell or WSL

ffmpeg Not Found

Ensure ffmpeg is installed and available in your PATH. You can test with:

ffmpeg -version

Whisper Model Download Failures

The first run downloads the Whisper model (base, ~150MB). Ensure you have internet connectivity and sufficient disk space.

Performance

Transcription speed depends on video length and CPU. For long videos, consider using a smaller model (edit script to use "tiny" or "small").

Customization

You can modify the script to use different Whisper models by changing the line:

model = whisper.load_model("base")  # Change to "tiny", "small", "medium", "large"

References

• OpenAI Whisper
• FFmpeg
• MoviePy