Claude Opus 5.5: Check Ad Audio Loudness Without Hiding Clipping

Use Claude Opus 5.5 to reconcile measured loudness, true peak and the delivery brief before increasing an ad's volume. A quiet integrated reading does not mean the entire mix can safely receive the same gain. Give the model measurements from the final mix and ask for a gain budget, then measure the exported file again.

Separate the three problems

A 2014 editor's question describes presenter recordings and background music that need different treatment. That is a historical self-report, not a tested preset. First decide which problem your ad has.

LUFS describes loudness; dBTP describes true peak. Neither proves speech intelligibility. The FFmpeg loudnorm documentation separates integrated loudness, loudness range and true-peak targets. Its linear mode has measurement and feasibility conditions; do not assume every requested target is achieved by a simple gain change.

Calculate the margin before making it louder

Consider a fictional mix measured at −22 LUFS with a −3 dBTP maximum. For this example only, the delivery brief requests −16 LUFS and a −1.5 dBTP ceiling. These are chosen teaching values, not a universal social-platform specification. The requested gain is +6 dB, which would project the peak to +3 dBTP under uniform gain. The peak ceiling permits only +1.5 dB, leaving a 4.5 dB gap between the two constraints.

Adding +1.5 dB would project loudness to roughly −20.5 LUFS for an unchanged gated measurement set. Actual remeasurement can differ. Do not label it a successful −16 LUFS delivery. Rebalance the mix or control peaks deliberately, then measure again. A limiter can change the sound; lowering a file after distortion occurred cannot reconstruct the lost waveform.

Copy the measurement-review prompt

Review an ad-audio delivery plan. Inputs: final-mix measurements, channel layout, duration, export format and the actual delivery brief. Do not claim to have listened or measured anything unless you used an audio tool.
For the illustrative values I=-22 LUFS and TP=-3 dBTP, calculate the gain requested by I=-16 and the gain permitted by a TP=-1.5 ceiling. Label all projected values as estimates. These targets belong to this example only.
Separate dialogue/music balance, integrated loudness, true peak and audible distortion. If the gain constraints conflict, propose a reversible mix adjustment or peak-control experiment; do not promise that normalization alone solves it.
Return a table with input measurement, target, calculation, proposed action and untested assumptions. Preserve a source copy. After export, require a fresh measurement of the actual delivered file and listening checks at the marked loudest and quietest passages. Mark absent evidence unknown.

This proposed Opus workflow follows the official guide's emphasis on explicit instructions. The arithmetic is reproducible; no audio file or model result was tested for this article.

Approve the file that will actually ship

Record the export filename, checksum, duration, channels, integrated loudness and maximum true peak. Listen to the opening, dialogue under music and the loudest transition on headphones and an ordinary small speaker. Keep before/after measurements tied to the same mix version. If the destination supplies a specification, use that current brief instead of copying this example's numbers.

**Can one normalization pass fix a buried voice?** It raises the music too. Correct the relative balance first.

**Does a safe true peak prove good sound?** No. Previously clipped audio, pumping or masked speech can remain.

Use approved product references in Panelly Studio for the accompanying creative. Review voiceover pronunciation separately: a clean level does not make a product name correct. This workflow does not imply a Panelly–Opus integration.

Sources and further reading

• Original reference · stackoverflow.com
• Original reference · ffmpeg.org
• Original reference · platform.claude.com