Quick answer: select the exact audio you intend to judge, choose Tools → Statistics, and copy only the fields required by the job. Use the complete programme for final LUFS and true-peak verification. Use a short, representative region for noise, DC offset, or fault diagnosis. Process only after you know which measurement failed, then run Statistics again on the same scope.
Statistics answers a narrow but important question: what's actually in the region you selected? It can locate sample extremes, report RMS and average value, estimate true peak, and measure programme loudness. It won't normalize, limit, repair, or export the audio.
Part of the Sound Forge mastering guides. For the processing sequence around these measurements, use the complete mastering workflow.
Run Statistics on the right region
The current Sound Forge 2026 Statistics documentation says the dialog reports the selected sound-file region. Keep that scope in mind when comparing reports: avoid measuring a short loud passage before processing and the full file afterward.
- Open the delivery source. Open the same sample rate, bit depth, and channel layout that you intend to approve or diagnose.
- Choose the analysis region. Select a noise-only pause for diagnosis, a problem passage for repair, or the complete programme for final loudness verification.
- Open Tools → Statistics. Sound Forge builds a read-only report for the selected sound-file region.
- Set the display formats. Choose Ruler Format and Level format only when another time or sample-level representation makes the report easier to read.
- Record the fields that answer the job. Copy integrated loudness, LRA, maximum true peak, RMS, average value, or sample positions according to the actual question and delivery brief.
- Copy the baseline. Use Copy to clipboard for the complete report, or select specific cells and press Ctrl+C.
- Process and measure the same region again. Apply a justified correction, render the result, and compare the same region with the baseline and named delivery specification.
A quiet diagnostic selection and a complete delivery selection can both be correct, but they answer different questions. Name the region in your notes. If the selection, channel layout, or loudness configuration changes between passes, the before-and-after numbers no longer describe the same test.
Four useful Statistics checks
Verify a final programme
Select the complete programme and record integrated loudness, maximum true peak, LRA, and the delivery method named in the brief. Do this on the rendered file rather than only on audio playing through an active effect chain. The render is the asset that will be encoded, uploaded, or handed to the client.
Measure a noise-only pause
Select several pauses that represent the background you actually need to control. Sample peaks help when setting a peak-detected gate. An RMS, mixed or sidechain detector can respond differently, so the highest sample is not a universal threshold formula. If one pause contains a breath or a passing vehicle, label it as an exception rather than allowing that event to define the whole recording's noise floor.
Investigate DC offset
Use Average value as a clue, then inspect the waveform around the zero line. Listen for associated clicks or distortion, not for DC itself. Compare more than one region when the recording changes source or hardware. A confirmed offset can affect headroom and some peak calculations, but an insignificant nonzero average isn't a reason to process an otherwise healthy file.
Prove what normalization changed
Copy a report before normalization, process the intended region, render, and copy the second report. Compare the requested target with the measured output. This catches selections that changed, processors that added gain later in the chain, and files that could not reach a loudness target without exceeding the peak limit.
Read the report without mixing up the fields
The report groups several measurement families in one place, and similar-looking negative numbers don't become equivalent just because they share a dialog. Decide whether you're locating a sample event, estimating average energy, checking peak safety, or measuring programme loudness before interpreting a value.
| Field group | Use it to answer | Do not infer |
|---|---|---|
| Minimum / maximum sample and position | Where the stored sample extremes occur | That every extreme is clipping or damage |
| RMS level | Average energy in a controlled region or an RMS-based brief | That RMS is interchangeable with LUFS |
| Average value | Whether DC offset may need closer inspection | That every nonzero result must be corrected |
| Zero crossings | A rough frequency clue for a very simple waveform | Pitch, edit safety, or frequency content of speech and music |
| Maximum true peak | Peak safety after waveform reconstruction | Perceived loudness or a universal delivery ceiling |
| Integrated / LRA / short-term / momentary | Overall loudness, variation, and loudest short windows | One universal LRA or streaming target |
Sample values and positions locate the event
Cursor, minimum, and maximum sample values describe stored samples and their locations. Jump to the reported position, zoom in, and listen. A clean transient, a clipped plateau, a click, and an intentional limited peak can all produce an extreme value, so the number is a locator rather than a diagnosis.
The same fields help with the Sound Forge Noise Gate workflow. Measure a representative noise-only pause when you need the highest noise event, but exclude breaths, chair movement, reverb tails, and music decay. Those sounds are programme material, not a stable noise floor. In the linked gate workflow, check the detector mode and listen to quiet wanted audio before committing the threshold.
RMS and integrated loudness answer different questions
Sound Forge defines RMS relative to the RMS value of a maximum-amplitude square wave. Its help also warns that one RMS value becomes less meaningful across a long selection with large volume variation. Use RMS when a client explicitly asks for RMS or when a controlled short region makes it useful. Do not convert a LUFS target into an invented RMS equivalent; the weighting, time behavior, and gating differ.
Integrated loudness is the gated programme measurement across the selected region. Maximum short-term uses three-second windows; maximum momentary uses 400-millisecond windows. Loudness Range describes variation in LU. EBU Tech 3343 does not impose one universal maximum LRA, so a broadcaster or delivery contract must supply any path-specific limit. For clips shorter than a minute, LRA is not a reliable acceptance measure: the underlying three-second windows provide too little material. Use the short-form measures requested by the recipient, such as maximum short-term loudness, instead.
Average value is a DC-offset clue
Average value is the sum of the samples divided by their count. A nonzero result can indicate DC offset, but a tiny deviation is not automatic permission to process the file. Inspect the waveform and confirm the fault before using the DC-offset correction workflow.
True peak is not the same as sample peak
Maximum true peak estimates the continuous waveform at a higher sample rate than the ordinary maximum-sample calculation. Its magnitude can therefore exceed the greatest absolute sample amplitude. For its filtered true-peak field, the help specifies the maximum of the filtered and unfiltered values. It is not simply a lower reading with DC removed; asymmetry and DC offset are reasons to inspect the result carefully.
The application help labels these values in dB FS, while EBU and ATSC delivery documents use dBTP. Record the label and value shown in the session, then compare them with the destination specification. Do not silently relabel an application readout or treat −1 dBTP as a universal ceiling.
Ruler Format and Level format change presentation only
Ruler Format controls how cursor and extreme positions are displayed. Level format controls how cursor, minimum, maximum, RMS, and average values are represented as samples, percent, or decibels. These choices change the display, not the audio samples. Pick the representation that matches your notes and keep it consistent when comparing reports.
Zero crossings are a narrow diagnostic
The zero-crossing count is the number of negative-to-positive crossings per second. It can suggest frequency for a simple periodic waveform, but speech, music, and noise contain many simultaneous components. Do not use this field as a pitch detector or as proof that an edit point will be click-free.
If the question is which frequencies are present, use the Sound Forge Spectrum Analysis guide instead. That view provides FFT, sonogram and snapshot comparisons; Statistics gives you a fixed numeric report.
Apply EBU R128 and ATSC A/85 to the correct delivery case
Sound Forge supplies measurements and presets; the standard or contract decides whether the file passes. EBU R 128 v5 sets programme loudness at −23.0 LUFS and states that true peak must not exceed −1 dBTP for linear audio during production. The same recommendation says different distribution systems and data-reduction rates may need a lower permitted maximum.
The production guidance gives −2 dBTP examples for MPEG-1 Layer II and Dolby AC-3. The measurement basis is ITU-R BS.1770-5, which remains in force. A valid PCM master can still need more headroom before encoding. The current BS.1770 recommendation does not override a delivery document’s specified measurement method, including the distinct ATSC dialogue path below.
The July 2026 ATSC A/85 Annex M quick reference lists −24 LKFS for content delivered without metadata when the parties have not agreed on another target, plus a maximum true peak of −2 dBTP. Its streaming guidance recommends selecting one consistent target between −23 and −27 LKFS unless the exchange parties agree otherwise. It also distinguishes dialogue-gated and full-programme measurement cases, so the phrase ‘ATSC is −24’ is not a complete delivery instruction.
| Reference | Loudness target | Maximum true peak | Scope that matters |
|---|---|---|---|
| EBU R 128 v5 | −23.0 LUFS | −1 dBTP | Generic linear audio during production; distribution can require more headroom |
| ATSC A/85:2026-07 default | −24 LKFS | −2 dBTP | Delivery without metadata when no other target has been agreed; follow the required dialogue or full-programme method |
| ATSC streaming guidance | One target from −23 to −27 LKFS | Confirm the delivery agreement | Select one consistent target unless the exchange parties agree otherwise |
Selecting the ATSC preset and reading ordinary Integrated LUFS is not proof of compliant long-form dialogue loudness. Annex M footnote 31 specifies BS.1770-1 without the relative-level gate added in later versions, combined with dialogue gating over the entire long-form composite mix. It does not mean measuring isolated dialogue stems or separate acts. Short-form uses full-programme BS.1770-3 or later. The Statistics help does not establish a dialogue-detection gate. For that delivery requirement, use the recipient’s specified meter and method; do not substitute this preset for them.
Use a before-and-after check that survives delivery
Save a baseline report before changing level. Record the filename, selected region, channel layout, loudness mode, and the fields required by the brief. Apply the smallest processing change that solves the failed measurement, render a new file, and measure the same scope again.
Simple gain arithmetic is a useful feasibility check. If a file measures −18 LUFS and about −2 dB true peak, reaching −14 LUFS with gain alone requires roughly +4 dB. That predicts a true peak near +2 dB before codec or sample-rate effects. The file lacks the headroom for that gain-only move. Change the loudness goal only if the brief allows it; otherwise consider controlled dynamics processing or a mix revision. This is a gain-only estimate, not a measured pass: gating, processing and export can change the final loudness result.
For the controls that change level, follow the Peak, RMS, and Loudness Normalize guide. A requested gain, limiter ceiling, or preset is an instruction to the processor, not proof of the rendered outcome.
Statistics, Loudness Meters, and Normalize have different jobs
Statistics creates a fixed, copyable report for a selected region. Loudness Meters show momentary, short-term, integrated, LRA, and true-peak behavior during playback. The current Loudness Meters help says those values reset when playback restarts. Normalize is a processing command that scans and changes level in its selected mode.
Use Statistics to document a before or final state, and use the meters to find where behavior changes over time. Neither tool is inherently more accurate merely because one is offline. The algorithm, configuration, measured region, channel handling, and delivery method still control the meaning of the result. The Sound Forge Loudness Meters guide covers the playback view in detail.
When two loudness readings disagree
Start by checking the test before blaming the meter. In Options → Preferences → Status, check the loudness settings; the current help uses both “Loudness Configuration” and “Loudness Modus” for the EBU/ATSC choice. A partial selection will not match the complete programme. A meter that was reset halfway through playback will not match a full-region Statistics report. Six-channel weighting, EBU versus ATSC mode, absolute LUFS versus relative LU display, and dialogue-gated versus full-programme methods can also change the result.
Next, confirm that both tools measured the same file. A live chain can differ from the rendered output, and lossy encoding can change peak behavior after the approved WAV leaves Sound Forge. Keep the baseline, the rendered report, and any post-encode QC result as separate records. If the numbers still disagree after scope and configuration match, document the exact versions and settings before deciding which measurement the receiving specification accepts.
Check the six-channel preference before comparing results
For six-channel files, choose Options → Preferences → Status and verify Enable surround processing for files with 6 channels. The documented measurement weighting adds about 1.5 dB to the left and right surround contributions when enabled; with it disabled, the help says all channels contribute equally. This is a measurement preference, not a command to raise the level of samples in the file. Two reports can differ when this preference is not matched. Six channels alone do not establish a correct 5.1 layout. Verify channel order and LFE treatment with a known channel fixture and a conforming reference before approving multichannel delivery.
Final Statistics checklist
- Measure the exact programme or diagnostic region required by the job.
- Record the Sound Forge version, loudness mode, channel layout, and six-channel preference.
- Compare integrated loudness with the named delivery target, not a generic streaming number.
- Compare maximum true peak with the production, codec, or distribution limit that actually applies.
- Treat LRA as a descriptor unless the destination publishes a limit.
- Use RMS only for an RMS-based brief or a controlled diagnostic region.
- Inspect reported sample positions before declaring clipping, a click, or DC offset.
- Keep the final report with the approved render.
Frequently Asked Questions
Where is Statistics in Sound Forge Pro 2026?
Select the intended audio and choose Tools → Statistics. The report covers the selected sound-file region and includes sample values and positions, RMS, average value, zero crossings, maximum true peak, integrated loudness, LRA, and maximum short-term and momentary loudness.
Does Statistics change or normalize the audio?
No. Statistics is a read-only analysis report. Use Normalize, Volume, a limiter, or another justified processor to change the audio, render the result, and then run Statistics again on the same region.
Does Statistics analyze a selection or the whole file?
It analyzes the selected sound-file region. Use a noise-only or problem region for diagnosis and the complete programme for final delivery verification. Keep the scope identical when comparing before and after reports.
What is the difference between sample peak and true peak?
Sample peak is the greatest absolute sample amplitude, whether the sample is positive or negative. True peak estimates the reconstructed waveform between samples, so its magnitude can exceed the sample peak. That matters for downstream conversion and encoding.
Is −1 dBTP the official EBU R128 maximum?
Yes for generic linear audio during production under EBU R 128. Distribution systems and data-reduced codecs can require a lower maximum; EBU production guidance gives −2 dBTP examples for MPEG-1 Layer II and Dolby AC-3.
What is the difference between Statistics, Loudness Meters, and Normalize?
Statistics creates a fixed report for a region. Loudness Meters show changing measurements during playback. Normalize scans and changes level. Measure first, process only when required, and verify the rendered result.
What are the current ATSC A/85 loudness and true-peak values?
ATSC A/85:2026-07 lists −24 LKFS for delivery without metadata when no other target is agreed and −2 dBTP maximum true peak. Its streaming guidance recommends one selected target between −23 and −27 LKFS unless the parties agree otherwise. Long-form dialogue and short-form full-programme measurement use different methods. Statistics’ ATSC preset alone does not establish the required dialogue-gated result; follow the recipient’s meter and delivery specification.
If a report points to a level problem, move to the mastering hub for the matching correction. If it points to noise, clicks, clipping, or DC offset, use the audio restoration hub before changing loudness.
Documentation checked: September 7, 2026. Sound Forge Pro 2026; EBU R 128 v5 and ATSC A/85:2026-07. Documented controls are not a substitute for testing the installed meter against the required delivery method.