How I built an AI assistant that picks my sound effects
I had a pack of 5,055 sound effects, music tracks and overlays, and I could never find the one I needed. Now I type one sentence, and Claude picks the sounds and places them in my editor. Here's how I put it together.
one sentence, no more digging
Tools used in this video
What you need
- A sound and music pack you own. Check its licence; mine is for my own videos only.
- Claude Code (included with Claude Pro). It writes and runs the small programs on your computer.
- Python and FFmpeg, both free. FFmpeg measures each file's length and loudness.
- An editor Claude can drive. I use Concat, which is free and open source.
- One afternoon for the first setup.
The steps
Put everything in one folder
All files in one place on your computer, one sub-folder per category: sound effects, music, overlays. Mine came to 5,055 files in 13 categories.
Build an index: one row per file
This is the assistant's memory. Each file gets a row: what it is, how long, how loud. Claude Code writes the program and FFmpeg measures the files. A real row from mine:
sfx / whooshfileswish-02.wavlength0.44 sloudness−17.2 LUFStagswhoosh, shortLUFS measures loudness. It tells Claude how far to turn a sound down so it never drowns out your voice.
Write a Claude skill for it
A skill is a short instruction file (
SKILL.md) Claude Code reads every time: where the index is, what to use where, how quietly music goes under the voice. An excerpt from mine:When an edit needs a sound, search the index. Whoosh before cards: shorter than 1.2 s, turned down to about −20 LUFS. Background music: calm, between −24 and −28 LUFS, under the voice. Never put anything from the "hooks" folder in a video; it's for ideas only.
Ask in one sentence
From here, you just say what you want: “Add a soft whoosh before every card and calm music underneath.” Claude searches the index, picks, sets the volume and places it in the editor.
You review and export
Claude never publishes anything for me. I watch the draft, change what I want, and only then export.
Where I got stuck
The download stopped at 244 files
Google Drive limits how often a public file can be downloaded. The fix was downloading each folder as a zip and copying from there.
The zip renamed a folder
“Animated Emoji's” came out as “Animated Emoji_s”, and the index couldn't find it. One rename fixed it.
Not all music is “free”
NCS and YouTube library tracks often ask for a credit and can still trigger Content ID claims. I pick from the corporate and library folders.
Try it: your first prompt
Paste this into Claude Code in the folder that holds your sounds. It only builds the index; it doesn't delete or move anything.
This folder holds audio and video files in sub-folders by category. Build an index (assets_index.csv): one row per file with these columns: category, file name, path, length (s), loudness (LUFS), transparency (for videos), and 3–5 tags from the file and folder names. Measure with ffprobe and ffmpeg's ebur128 filter. Don't delete, rename or move any file. At the end, print how many files you found per category.