- Home
- Skills
- Game Development
- Unify an Audio Folder: same loudness, shared sound (FFmpeg)
More screenshots
Works with the AI Agents you already use
Unify an Audio Folder: same loudness, shared sound (FFmpeg)
Ask an agent to "normalize this folder" and it runs `loudnorm` on every file.
$3.99
Unify an Audio Folder: same loudness, shared sound (FFmpeg)
Example session with this skill installed
I've got a folder of game sounds in game_sfx, grabbed from all over the place. Some are super loud, some I can barely hear. Can you make them consistent? I'd like them to feel kind of retro too, like an old console game.
- Read your context and instructions
- Compiled the unify an audio
Scan + audition. The agent scans the folder and renders an audition, then stops:
coin_pickup.wav: 0.42 s, 44100 Hz, 1 ch, -1.1 dBFS loudest
enemy_spotted.wav: 1.99 s, 24000 Hz, 1 ch, -17.5 dBFS loudest
explosion_far.ogg: 1.80 s, 44100 Hz, 2 ch, -40.0 dBFS loudest QUIET
jump.wav: 0.24 s, 22050 Hz, 1 ch, -6.9 dBFS loudest
laser_zap.mp3: 0.43 s, 48000 Hz, 2 ch, -18.5 dBFS loudest
menu_click.flac: 0.09 s, 48000 Hz, 1 ch, -31.8 dBFS loudest QUIET
objective_complete.mp3: 2.95 s, 24000 Hz, 1 ch, -27.5 dBFS loudest
Lengths: 0.09 to 2.95 s, median 0.43 s; 7 under 3 s (those always use the sound-effect method)
"I've put together an audition reel so you can pick the retro character by ear before I run it over the whole folder. Files: game_sfx_audition/audition_reel.wav (8 takes, a beep before each) and audition_cues.txt. It compares your clean coin pickup and the far-off explosion with no effect, then through three retro rungs:
**nes**, harshest, thin buzzy 8-bit crunch;
**snes**, strong, muffled crunch;
ps1, grainy and roomy, a middle ground. Have a listen and tell me which rung feels right (or n64/saturn for milder, dreamier distortion)."
Feedback: "PS1 sounds right. Go ahead with the whole folder."
Hand-back: "All 7 files processed successfully, no failures. They're now in game_sfx/unified/, each named <original>_sfx_ps1.<ext>, same formats as the originals, loudness leveled, and all run through the PS1 grain/room character. Your originals in game_sfx are untouched. Next steps if you want them: trim silence at the edges, or convert the whole set to one format (e.g. OGG) for your engine."
Connects securely to your tools. The creator never sees your data.
About this skill
Ask an agent to "normalize this folder" and it runs loudnorm on every file. For a folder of game sounds that's the wrong tool: integrated loudness needs seconds of material and ignores quiet parts, so a 200 ms blip measures as near-silence and gets boosted into clipping. This skill gives your agent the method that works, tuned by ear over several listening rounds and checked on a 600-file batch, plus an optional shared texture so sounds from different sources finally feel like one set.
What it does
- Matches loudness the right way for the material. Four categories, each with its own measuring method and target: sound effects (RMS peak -13 dBFS, one static gain, -1 dBTP ceiling), voice and speech (two-pass loudnorm, -16 LUFS), music (-14 LUFS) and ambience (-20 LUFS, wider range). Anything under 3 seconds always uses the sound-effect method. The categories never EQ or compress, so a wrong pick can't damage a sound.
- Gives the folder one shared sound, if you want it. A six-rung quality ladder, from harsh 8-bit crunch to a light fizzy grain, with console flavours as a strength yardstick (NES, SNES, PS1, N64, Saturn, PS2). Degrading every file to the same level is what makes mismatched sources match. Or a "coming from" effect: radio, telephone, megaphone, underwater, cassette tape, vinyl, another room, gritty, brutal. Or a polish: podcast, punchy SFX, warm, bright. Every effect runs before the loudness match, so the files still match afterwards.
- Trims silence: at the ends only, or long pauses throughout, with a threshold that follows each file's level so quiet recordings keep their soft passages.
- Finds the odd ones out. The scan flags files 10 dB or more under the folder's median, and can process just those.
- Leaves the originals alone. Output goes to a mirrored folder with clear names (
door_sfx_ps1.wav), keeps each file's format, rate and channels unless you ask for another format, and re-runs overwrite their own output instead of stacking.
How it works
- Scan. One command lists every file: length, rate, channels, loudest moment, quiet flags, unreadable files.
- Category. The agent proposes one from the lengths and names (short game sounds, voice lines, music, ambience) and splits mixed folders into one run per kind.
- Audition (when an effect is in play). Your cleanest and roughest file through 2–4 candidate effects, joined into one reel with a beep before each take and a cue sheet. You pick by ear.
- Run. The whole folder in one pass, one log line per file: the gain applied, notes, failures with a reason.
What it fixes that improvised loops get wrong: short sounds blown up into clipping by loudnorm, effects that shift the loudness after it was matched, volumedetect reading float overs as 0 dB, loudnorm rejecting its own first-pass values on clipped files, loudnorm silently writing 192 kHz files, cover-art streams breaking the encode, FFmpeg's vibrato and bitcrusher writing NaN samples, and two source files colliding on one output name.
What's inside
- A paved-path SKILL.md
- Five references: the four categories and why, the effects with their exact chains and what each sounds like, silence trimming, the pipeline (order of operations, naming, folder rules, doing it by hand), and a failure log
- One self-contained Python script (
scan,run,audition,effects); numpy only for silence trimming
Where it stops: it levels and textures, it doesn't repair. A quality rung can't make a bad file good, and the console rungs are a strength scale with a bit of flavour, not chip emulations. For voice characters (demons, robots, radio chatter at three intensities), the Voice FX skills are the better fit; this one is the last step that makes a whole folder sit together.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Game Development.
Needs FFmpeg and Python on the machine where the agent runs. It levels and textures audio; it doesn't repair noise, clipping or bad recordings. Spiky sounds can land a little under target, because the true-peak ceiling wins and no limiter is added (that would reshape the sound). Effects are one strength each; the ladder's console names are a severity scale, not emulations. Silence trimming needs numpy. Offline batch processing, about a second per file with an effect and trimming.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed today
- Software
- FFmpeg with ffprobe on PATH
Python 3.10+
- Optional
- numpy (silence trimming only)
Loudness matching and effects need only FFmpeg and Python; `pip install numpy` only if you want silence trimmed.