Does a wider visible range automatically mean better sound?

No. A wider visible range is not automatically a quality verdict. The chart shows output behaviour against this reference sample, not listener preference or transparency.

Are the boost response charts raw FFT exports?

No. They are extracted from the DeltaWave aligned-spectrum plots for the original-WAV comparisons. They are useful for comparing the tested ABR builds, but they remain extracted chart data rather than direct raw FFT exports.

ABR Frequency Response Lab

Original WAV comparison · LAME 3.101 and 4.0 · ABR boost

How do the ABR builds compare against the original WAV?

This page uses the fresh DeltaWave aligned-spectrum reports for official LAME 3.101, official LAME 4.0, and the current MinGW and Visual Studio cVBRb builds. Both unboosted and ABR-boosted outputs are compared with the same original WAV.

It is a technical lab view rather than a listening test. The charts show upper-band roll-off and the limited response changes introduced by the current ABR boost mode.

View ABR charts View ABR metrics View ABR speed/size Read method

The short version

Official 3.101, official 4.0 and the unboosted cVBRb builds are effectively aligned in the extracted upper-band response for this sample.

Both boosted builds move the 12 dB crossing from about 16.75 kHz to 17.50 kHz, while the 3 dB and 6 dB crossings remain unchanged.

The boosted outputs are slightly larger, with encoding speed remaining close to the corresponding unboosted builds.

This is not a listening-test verdict. These are extracted DeltaWave aligned-spectrum curves, useful for illustrating behaviour against the original WAV rather than proving subjective transparency.

Current ABR response and boost comparison

The ABR report set includes a single boost variant. This section applies the same aligned-spectrum extraction and derived-metric approach used for the VBR/cVBRb frequency-response page.

Official LAME 3.101 and current LAME 4.0 unboosted ABR response against the original WAV — Official 3.101, official 4.0 and the two unboosted cVBRb builds. Their extracted response curves are nearly coincident in this test.

Current LAME 3.101 and 4.0 ABR frequency response comparison including both boosted builds — Current ABR overview. This chart includes all six fresh reports: official 3.101 and 4.0, both unboosted cVBRb builds and both boosted cVBRb builds.

MinGW and Visual Studio ABR boost compared with their unboosted cVBRb outputs — ABR boost versus the corresponding unboosted MinGW and Visual Studio cVBRb outputs. The boost extends the 12 dB crossing in both builds.

Deviation from original WAV for modern ABR rows — Focused view of the roll-off region. Lower curves indicate less loss against the original WAV.

ABR grouped bar chart showing 3 dB, 6 dB, and 12 dB drop points — Current ABR drop-point summary for all six fresh reports.

ABR bar chart showing average loss versus the original WAV in the 18 to 21 kilohertz band — Average ABR loss in 18–21 kHz for the six current reports. The boost changes this average only slightly, while extending the later 12 dB crossing.

ABR derived response metrics

Series	3 dB drop	6 dB drop	12 dB drop	Mean loss 18–21 kHz	Mean loss 19–22 kHz
3.101 official	16.00 kHz	16.03 kHz	16.75 kHz	35.43 dB	39.10 dB
4.0 official	16.00 kHz	16.03 kHz	16.75 kHz	35.43 dB	39.10 dB
4.0 cVBRb MinGW	16.00 kHz	16.03 kHz	16.75 kHz	35.43 dB	39.10 dB
4.0 cVBRb Visual Studio	16.00 kHz	16.03 kHz	16.75 kHz	35.43 dB	39.10 dB
4.0 cVBRb MinGW boost	16.00 kHz	16.03 kHz	17.50 kHz	35.11 dB	39.18 dB
4.0 cVBRb Visual Studio boost	16.00 kHz	16.03 kHz	17.50 kHz	35.11 dB	39.19 dB

These ABR metrics are derived from extracted aligned-spectrum curves, using the same approach as the VBR/cVBRb frequency-response page.

Archived-release ABR context

The current 4.0 results remain the main comparison. Earlier releases are included here because several show distinct measured behaviour.

Archived-release ABR frequency-response comparison from LAME 3.95.1 through 3.100 — Archived-release ABR upper-band response from LAME 3.95.1 through 3.100.

Version	3 dB drop	6 dB drop	12 dB drop	Mean loss 18–21 kHz
3.95.1	16.03 kHz	16.05 kHz	16.78 kHz	35.87 dB
3.96.1	16.03 kHz	16.18 kHz	18.00 kHz	36.71 dB
3.97	16.00 kHz	16.03 kHz	16.05 kHz	36.78 dB
3.98.4	16.00 kHz	16.03 kHz	16.08 kHz	36.86 dB
3.99.5	16.00 kHz	16.03 kHz	16.75 kHz	35.43 dB
3.100	16.00 kHz	16.03 kHz	16.75 kHz	35.43 dB

3.96.1

Shows the latest 12 dB crossing in the archived-release set, while also producing the strongest residual measurements on the companion closeness page.

3.98.4

Represents the efficiency-focused period: compact output and relatively fast encoding, with a more aggressive upper-band roll-off.

3.99.5

Combines the fastest archived-release encoding result with the same extracted response profile later measured for 3.100.

3.100

Matches 3.99.5 in output size and extracted response here, but is substantially slower in the submitted benchmark.

ABR output-size and speed summary

Series	MiB	Elapsed seconds	MiB/s
3.101 official	12.982	5.770	2.250
4.0 official	12.982	5.740	2.262
4.0 cVBRb MinGW	12.982	5.870	2.212
4.0 cVBRb Visual Studio	12.982	5.450	2.382
4.0 cVBRb MinGW boost	13.339	5.910	2.257
4.0 cVBRb Visual Studio boost	13.339	5.460	2.443

This table uses the current ABR benchmark CSVs for official 3.101 and 4.0, the two unboosted cVBRb builds and both boosted outputs.

Method notes and limitations

The ABR response charts on this page are extracted from the embedded DeltaWave aligned-spectrum SVG plots in the ABR report set. The dashed reference line is the original WAV, while the solid lines are decoded ABR outputs.

The charts use the IBM Carbon data-visualization palette for categorical series, with a neutral dashed reference line for the original WAV.

The derived metrics include 3 dB / 6 dB / 12 dB drop points and average loss across the 18–21 kHz and 19–22 kHz bands. These are approximate because they come from extracted plot geometry rather than direct FFT exports.

ABR has one boost mode. MinGW and Visual Studio results are retained as separate rows because they are separate build toolchains, even where their response measurements are nearly identical.

FAQ

Is this a precise frequency-response measurement?

No. The charts are extracted from DeltaWave aligned-spectrum images, not from raw FFT exports. They are useful for comparing ABR behaviour, but they are still extracted visual data rather than lab-grade response measurements.

Why is the ABR drop-point chart scaled differently?

The ABR rows diverge much earlier in the upper band than the VBR/cVBRb rows on this sample, so the chart needs a lower y-axis range to avoid clipping the bars.

How many ABR boost levels are included?

One ABR boost mode is included, measured separately for the MinGW and Visual Studio cVBRb builds.