Day 03 · May 2, 2026

Studio-quality iPhone audio

Cleanvoice · audio polish · locked edit

"I recorded this on my iPhone. Then Claude made it sound like a studio mic."

Day 3 turns phone audio into a repeatable agent step: extract the voice track, send it to Cleanvoice, and mux the cleaned audio back into the finished edit without rerendering the picture.

You don't need to read this

If you have Claude Code, install the Day 3 plugin and ask Claude to clean up a Reel's voice audio with Cleanvoice.

$ /plugin marketplace add andrewjiang/andrewislearning

$ /plugin install day-03-cleanvoice-audio@andrewislearning

What the plugin does

The skill is intentionally narrow: improve the voice without disturbing timing, captions, or the approved video render.

Extract voice

Pulls the spoken audio into a clean WAV handoff file.

Cleanvoice edit

Runs noise removal, studio polish, and loudness normalization.

Preserve proof

Keeps deliberate raw-before samples untouched when the demo needs contrast.

Mux, don't rerender

Copies the video stream and swaps only the audio track.

Compare

Checks for clipped words, doubled consonants, and timing drift.

Master

Keeps voice clear under captions, SFX, and quiet music beds.

The guide

Set your API key, then let Claude run the audio pass after the edit is locked.

$ export CLEANVOICE_API_KEY="..."

$ python3 pipeline/enhance_audio.py input.mp4 --provider cleanvoice --out output.voice.mp4

Behind the scenes, the flow is upload, create edit, poll status, download the cleaned audio, and mux it back into the video. The important part is that this stays an audio-only pass.

The script

Day 3 used the before/after audio demo as the hook, then explained the series rule after the proof beat.

Read the final spoken script

Hook

I recorded this on my iPhone. Then Claude made it sound like a studio mic.

Before / after demo

This is what it sounds like before. And this is what it sounds like after.

Series rule

This is Day 3 of teaching AI to be my video editor. I can only record with my phone and laptop. Everything after that has to be handled by an AI agent.

Cleanvoice flow

To improve the audio from my iPhone recordings, my AI agent sends the raw voice track to Cleanvoice and gets back studio-quality audio a few seconds later.

Loop

And that's why, for Day 3, I built...

The source

github repo plugin source full guide Cleanvoice API docs

Next - Day 04

AI-assisted transitions using a first frame, a last frame, and shot planning.