kutter

a cli program that can split and organize mkv files

why not just use mkvtoolnix?

I am actually! this is just a automated workflow on top of mkvtoolnix and ffmpeg that I can plug into tdar to make my life easier when ripping shows. that and I thought it'd just be a fun project to flex my programmer brain.

goal for this project

• automate the process of detecting episodes and where to split
• integrate into tdarr flow to split files before/during/after encoding
• organize the files into a format that can just plug directly into jellyfin/plex

strategies to use to find episodes

Start with duration matching as baseline. Adjust using chapter count expectations. Use chapter title metadata or duration anomalies (e.g. intros/outros) to fine-tune.

cumulative duration matching

Build a cumulative chapter time list.

Iterate over the chapters and keep summing durations until the total matches the average episode duration.

If a chapter ends slightly before or after the target, use a tolerance window

This is often the most accurate for episodes with uniform pacing.

Pros: Most direct match to episode duration Cons: Breaks if pacing is uneven or some episodes have long intros/outros

duration anomaly detection

When to use: Chapter titles are absent or unreliable, but durations are available.

What to do: Calculate the mean and standard deviation of chapter durations.

Flag chapters that are:

Much shorter (e.g. <10s, common for intros/outros)

Much longer (could signal the start of a new episode)

Identify consistent long-duration clusters (near your average episode length).

Use these chapter indices as candidate split points.

Why it helps: Some releases use chapters inconsistently, but episode breaks often fall near unusually long chapters (i.e., full content between intros/outros).

sliding window alignment

If average episode durations are reliable, use a sliding window over the chapter durations:

• Start at 0, and at each step, look ahead over n chapters.
• Find the window where sum(chapter_durations) ≈ average_episode_duration ± tolerance.

Pick the chapter boundary closest to that sum.

This is similar to what you already do, but allows you to adjust dynamically when previous episodes were under/over time.