Versely
    Guides

    This loudness meter is a BS.1770-style estimate, not EBU R128

    It K-weights and gates in JavaScript with no true-peak oversampling, so broadcast delivery still needs Youlean, Nugen or ffmpeg ebur128.

    Versely Team••6 min read

    It K-weights and gates in JavaScript with no true-peak oversampling, so broadcast delivery still needs Youlean, Nugen or ffmpeg ebur128.

    Loudness Meter is labelled an estimate on the tool itself. Peak dBFS, whole-file RMS, and a BS.1770-style integrated figure with K-weighting and gating. That is a sanity check in this tab. It is not a certified EBU R128 meter, and it will not stamp a delivery.

    People still screenshot the LUFS number and paste it into a spec sheet. Do not. The number is useful. The stamp is a different product.

    What the tab actually computes

    Three readouts, plus the metadata that tells you what was decoded:

    • Peak — the single loudest sample, in dBFS. Sample peak, not true peak. A click or a plosive can hit 0 dBFS while the file still sounds quiet.
    • Average RMS — energy across every sample in every channel. Closer to a crude mean-square than to perceived loudness.
    • Integrated (est.) — K-weighted, 400 ms blocks, 100 ms hop, absolute gate then relative gate, then a mean of the survivors converted with the BS.1770 offset (−0.691 + 10 log10 of mean-square). The UI reports how many blocks survived gating against how many were measured, plus duration, sample rate, and channel count.

    K-weighting here is the two-stage idea from ITU-R BS.1770: a high-shelf around 1.7 kHz, then an RLB high-pass, bilinear-transformed to whatever sample rate the Web Audio decoder produced. Surround-style channel weights apply after that (1.0 on the first two channels, 1.41 on the rest). The absolute gate is the −70 LUFS floor. The relative gate is 10 dB below the ungated mean of blocks that cleared the floor.

    That is the BS.1770 idea, running in JavaScript on a buffer the browser already resampled. It is not a certified implementation, and it does not add the rest of EBU R128.

    What EBU R128 still requires

    EBU R128 is a delivery recommendation that sits on top of BS.1770. Broadcast specs care about more than one gated integrated number:

    • a target integrated loudness (classically −23 LUFS for EBU programme delivery; streaming platforms often want something hotter, commonly around −16, as their spec, not this tool’s)
    • a maximum true-peak, which is not the same as sample peak
    • often a maximum short-term or momentary loudness, which this tab does not compute
    • a meter that was built and verified against the recommendation, not a biquad in a blog-adjacent utility

    This tool does not compute short-term. It does not compute momentary. It does not dialog-gate. It does not oversample for true-peak. It does not emit a loudness range. The decoder has already resampled to the AudioContext rate — often 48 kHz — so even the sample-peak is a peak of a resampled buffer, not of the bitstream you will hand to a broadcaster.

    Use Youlean, Nugen, or ffmpeg ebur128 when the file has to pass a spec. Use this tab when you want to know whether a mix is obviously hot, obviously quiet, or peak-limited while still reading soft.

    True-peak is the hole

    Sample peak is the largest absolute value in the PCM you already have. True-peak is what happens when that PCM is reconstructed — typically estimated by oversampling 4× and watching for inter-sample peaks that sit above 0 dBFS even when no sample does.

    Those inter-sample peaks are why a file that “never clips in the editor” still distorts after a codec or a DAC. EBU R128 cares about that ceiling. This meter cannot see it. There is no oversampling stage. Peak 0.0 dBFS here means a sample hit full scale, not that the reconstructed waveform stayed legal.

    That is also why a delivery house will bounce you for true-peak even when your integrated estimate looked fine. The two numbers answer different questions. Integrated (est.) is “how loud does this feel, roughly, after K-weighting and gates.” Peak is “did a sample clip.” True-peak is “will it clip after reconstruction.” Only the first two are in this tab, and the first is an estimate.

    The file is capped at 80 MB and 20 minutes once decoded. Compressed podcasts expand into PCM in memory. That guard is local, not a credit gate. Nothing is uploaded. The work stays on the device.

    When the estimate is enough

    It is enough when the job is “is this export in the right neighbourhood before I spend a generate.” A native-audio clip from VEO 3.1 can be peak-limited and still quiet; that pairing is the point of showing peak next to integrated. If the estimate is wildly off your platform’s target, fix the mix in an editor or a limiter. Do not prompt a video row to “hit −16.” The model will invent a mix. It will not give you a meter.

    What it costs is the hub for jobs that spend credits. Attaching a separate voiceover is a different meter — per 1,000 characters plus a flat attach — documented as voiceover for a script and as the additive route on AI video with sound. Measuring the result afterwards is still this tab.

    If you needed to cut the file rather than measure it, Audio Trimmer is the sibling. It re-encodes to 16-bit PCM WAV. It is also not a delivery path.

    Disconnect the network. The meter still runs. Do not put its LUFS on a delivery memo.

    FAQ

    Is this EBU R128? Can I use it for broadcast delivery?

    No. It follows the BS.1770 idea — K-weighting, 400 ms blocks, absolute and relative gates — but it is not a certified meter. There is no true-peak oversampling, the decoder has already resampled, and the filters run in JavaScript. Use Youlean, Nugen, or ffmpeg ebur128 for a delivery spec.

    Why is my peak 0 dBFS when the file still sounds quiet?

    Peak is the single loudest sample. A click or a plosive can hit 0 dBFS while RMS and integrated loudness stay low. Loudness is closer to perceived level; peak is what clips. A file can be both peak-limited and quiet.

    Does the integrated number include true-peak or short-term loudness?

    No. Integrated (est.) is a gated mean-square after K-weighting. Short-term, momentary, loudness range, and true-peak are not computed. If a spec names those, this tab cannot answer them.

    Do my files get uploaded anywhere?

    No. Browser, Web Audio, local memory. Nothing is sent to a server, which is also why it keeps working with the network off. Generation on Versely is the other meter, and it always costs credits.