PRACTICAL GUIDE

How to read the audio analyzer

Learn to read loudness, peaks, stereo and format. Use measurements to investigate your audio without treating them as a quality score.

On this page11 sections

Open the analyzer and choose your file or Try a piano example. Playback and measurement use the same decoded audio; the original stays unchanged.

  1. Listen, then keep Whole track or choose an excerpt. Start and End define what is measured; the white cursor only moves playback.
  2. Choose Analyze track or Analyze excerpt. Review Levels, Frequency and Stereo. Charts start at zero for the measured interval.
  3. Export report opens the measurements and charts in another tab. Print or save as PDF there, or download TXT or CSV. Reports contain no audio.

On a phone, completed analysis takes you to the results. Back to audio returns to the controls. Editing the interval clears the previous results and report; canceling lets you retry with the prepared audio.

Loudness: how loud it sounds to the ear

For a first pass, load the piano and choose “Analyze track”. Then select an excerpt with a clear attack and analyze it again. Compare the peak and loudness of each interval before looking at the spectrogram.

Sample peak describes amplitude; LUFS estimates loudness through weighting and time analysis. The integrated reading summarizes the file with gates that exclude very low blocks. It does not measure artistic quality or predict playback on every system.

Momentary loudness uses 400 ms windows and short-term loudness uses 3 s. Loudness range (LRA) describes the variation of the latter through its 10th and 95th percentiles after its own gating. A short excerpt provides little context; a low value does not prove a compressor was used. Compare the same passage or the same complete song.

Peak, true peak and clipping: how close you got to the digital ceiling

Sample peak, in dBFS, is the largest absolute sample amplitude. In integer PCM, 0 dBFS is the full-scale reference. A value near that ceiling cannot establish whether a file was mastered or distorted. Floating-point files and processing can represent values above that reference; an output device or later conversion may still clip them.

Estimated true peak, in dBTP, accounts for peaks between samples through oversampling. HertzKit provides an estimate, not certification. Its full-scale sample counter uses a 0.9999 threshold and also counts runs of three or more: it flags places to review, but does not by itself prove audible clipping. DC offset is the signal average; it may consume headroom and cause clicks during editing.

RMS, crest factor and loudness range: three ways of looking at dynamics

RMS describes effective amplitude. Crest factor compares peak with that RMS: an ideal sine has about 3.01 dB of crest. A small factor may come from limiting, distortion or the sound itself; it does not establish how the file was produced. Constant gain raises peak and RMS equally and preserves crest factor while no clipping occurs.

LRA observes changes in loudness over time. It differs from crest factor: a sound can have modest attacks yet move from a quiet section to a loud one. Interpret both alongside listening and the arrangement.

Stereo: correlation, balance, width and the goniometer

In stereo, correlation describes the relationship between channels. Near +1 they can be proportional with the same polarity; near 0 there is little linear relationship; negative values are a reason to check the mono sum. Correlation of −1 only means complete cancellation when amplitudes are equal too. Balance compares left/right energy and width compares side with mid energy.

The goniometer visualizes these relationships: equal signals in both channels form a vertical line; equal signals with opposite polarity form a horizontal one. It is a visual aid. Global values can hide a brief effect, so listen to the passage that concerns you before changing it.

The average spectrum and the octave-band balance

The cyan line summarizes spectral power across analyzed windows and the amber line shows their maxima. The frequency axis is logarithmic. Levels are calibrated to help read tones; a sine between FFT bins can appear lower than one exactly on a bin. Do not treat one graph reading as an exact amplitude measurement.

The ten bands, from 31.5 Hz to 16 kHz, group that energy to show which areas dominate. They are estimates over analysis windows and frequency ranges, not a sum that must equal the global RMS exactly. The end bands have resolution and coverage limitations.

In HertzKit, the spectrum, bands and spectrogram start from the mono average (L + R) / 2. Opposing stereo content can be reduced or disappear there while channel RMS remains high. Compare these graphs with stereo playback and the mono sum. A flatter curve is not a better mix.

The spectrogram: why it's on a logarithmic scale

The spectrogram places time horizontally, frequency vertically and energy in brightness. A logarithmic scale gives each octave equal space and makes bass regions easier to distinguish. Hover to inspect time, frequency and level.

A persistent line near 50 or 60 Hz may suggest hum, but it can also be part of the music. A high-frequency roll-off can come from the content, a filter or earlier encoding: the image alone cannot identify the cause or file quality. Listen and compare with the original.

Dominant frequency: what it is, and more importantly, what it isn't

This estimates the dominant peak in the average spectrum, not the song’s key or a transcription of its melody. A bass, kick or harmonic may dominate the result. The accompanying note and cents convert those Hz relative to a reference.

For a sustained microphone note, use the tuner; for a calculated equivalence, use the frequency-to-note converter. In stereo, remember that this spectrum comes from the mono sum and can lose content that cancels.

Format, sample rate and bit depth: what the file says about itself

The analyzer tries to read format and sample rate from file bytes. Browser decoding may resample to the audio context’s rate; HertzKit therefore tries to analyze at the original rate when possible and shows “Analyzed at” when it differs. This context does not necessarily have the same rate as your sound device.

In integer PCM, bit depth describes sample precision. Floating point uses another representation and provides processing headroom: “32-bit float” does not mean 32 bits of useful recorded detail. For formats such as MP3 or AAC, bitrate and codec describe encoding; they are not interpreted as a PCM bit depth. Ogg is a container that can hold different codecs.

When available, “effective bits” examines sample patterns. A 24-bit WAV whose bottom eight bits are always zero is consistent with shifted 16-bit values, but does not establish recording origin or certify quality, noise or converter resolution. Gain, processing and dither can change that pattern. The analyzer also cannot reliably determine whether a file received dither. For formats whose metadata it cannot read, it indicates the rate used for analysis.

A concrete use: checking a file before sending it off for mastering

Check duration, channels and requested format. Review estimated true peak, full-scale areas and unexpected mono losses. Ask the recipient which headroom and format they need: no single peak, correlation or crest value approves every delivery.

Listen to the start, end and demanding passages. The report helps locate things to review; it does not verify a recording’s “real 24 bits” or replace listening. Keep the original and compare at matched loudness before deciding. Continue with gain staging or compare two versions.

Technical references: EBU Loudness, EBU Tech 3342, MDN: audio decoding and Audacity: bit depth. HertzKit details describe this tool’s scope, without certification or affiliation.

Turn a reading into a decision

The first reading summarizes loudness, estimated true peak and full-scale samples. All measurements remain in the expandable section. To check a change, use the Delivery review and A/B comparison and compare the same passage at matched loudness.