Average-bitrate testing confirms the performance story
This page compares LAME ABR encoding at 128, 160, 192, 224, 256 and 310 kbps. ABR is target-bitrate driven, so it gives a useful contrast to the quality-based VBR curve study.
LAME ABR Performance Study
The short version
ABR keeps the same performance pattern. The optimized 3.101b3 cVBRb build is fastest at every tested ABR target.
3.100 again matches 3.99.5 output byte-for-byte. It is still much slower, so the slowdown is not only a VBR-path issue.
ABR remains useful for bitrate-budgeted material. It was historically attractive for AVI/movie audio and spoken-word content because it can hold bitrate down during quiet passages while still targeting a known average.
ABR performance charts
These charts use LAME 3.99.5 as the baseline. A value of 1.0x means equal speed; larger values mean faster encoding on the test system. Output-size differences near 0% mean the file size was effectively unchanged for that ABR target.
These charts compare output size and encoding time only. They do not claim audible quality differences.
Why ABR is being tested
ABR sits between CBR and quality-based VBR. It aims for a chosen average bitrate, but it can still use different frame sizes, which made it useful when final file size mattered more than a pure quality target.
Historically, that made ABR attractive for AVI movie audio, dialogue-heavy soundtracks, spoken word and audiobooks. Quiet passages, ambience and pauses can use fewer bits for long periods, leaving more room for complex moments without abandoning the target average bitrate.
ABR headline results
Each row is one ABR target. The optimized 3.101b3 builds are Visual Studio 2026 builds; all other rows are MinGW plain builds.
| Target | Smallest output | Fastest plain build | Fastest overall |
|---|---|---|---|
128 kbps |
3.98.4 52.386 MiB |
3.99.5 39.86 s |
3.101b3 cVBRb Optimized 34.93 s |
160 kbps |
3.98.4 65.682 MiB |
3.99.5 40.01 s |
3.101b3 cVBRb Optimized 35.80 s |
192 kbps |
3.101b3 cVBR 79.441 MiB |
3.99.5 38.33 s |
3.101b3 cVBRb Optimized 34.96 s |
224 kbps |
3.101b3 cVBR 93.050 MiB |
3.99.5 39.34 s |
3.101b3 cVBRb Optimized 35.66 s |
256 kbps |
3.101b3 cVBR 106.432 MiB |
3.99.5 40.54 s |
3.101b3 cVBRb Optimized 36.75 s |
310 kbps |
3.97 124.895 MiB |
3.99.5 40.98 s |
3.101b3 cVBRb Optimized 36.56 s |
Average encode time by ABR target
Lower is faster. Times are elapsed seconds for the tested album set.
| Version / build | Toolchain | 128 | 160 | 192 | 224 | 256 | 310 |
|---|---|---|---|---|---|---|---|
| 3.95.1 | MinGW plain | 48.22 s | 48.33 s | 49.13 s | 46.60 s | 47.32 s | 47.10 s |
| 3.96.1 | MinGW plain | 52.70 s | 49.77 s | 47.42 s | 46.86 s | 48.45 s | 48.33 s |
| 3.97 | MinGW plain | 52.80 s | 54.02 s | 51.64 s | 52.08 s | 53.33 s | 55.02 s |
| 3.98.4 | MinGW plain | 42.22 s | 43.42 s | 41.05 s | 41.87 s | 42.84 s | 43.95 s |
| 3.99.5 | MinGW plain | 39.86 s | 40.01 s | 38.33 s | 39.34 s | 40.54 s | 40.98 s |
| 3.100 | MinGW plain | 73.53 s | 75.37 s | 73.34 s | 73.16 s | 74.57 s | 75.24 s |
| 3.101b3 cVBR | MinGW plain | 59.01 s | 61.02 s | 59.14 s | 60.25 s | 61.18 s | 61.01 s |
| 3.101b3 cVBR Optimized | Visual Studio 2026 | 35.53 s | 37.06 s | 36.74 s | 38.26 s | 39.19 s | 39.75 s |
| 3.101b3 cVBRb Optimized | Visual Studio 2026 | 34.93 s | 35.80 s | 34.96 s | 35.66 s | 36.75 s | 36.56 s |
Output size by ABR target
Lower numbers mean smaller files. Because ABR targets average bitrate, these rows converge more tightly than the VBR curve data.
| Version / build | Toolchain | 128 | 160 | 192 | 224 | 256 | 310 |
|---|---|---|---|---|---|---|---|
| 3.95.1 | MinGW plain | 55.994 MiB | 70.825 MiB | 86.075 MiB | 100.384 MiB | 113.605 MiB | 131.668 MiB |
| 3.96.1 | MinGW plain | 55.780 MiB | 70.634 MiB | 85.375 MiB | 99.433 MiB | 112.649 MiB | 131.205 MiB |
| 3.97 | MinGW plain | 52.398 MiB | 65.717 MiB | 79.448 MiB | 93.078 MiB | 106.868 MiB | 124.895 MiB |
| 3.98.4 | MinGW plain | 52.386 MiB | 65.682 MiB | 79.445 MiB | 93.310 MiB | 107.302 MiB | 125.839 MiB |
| 3.99.5 | MinGW plain | 53.979 MiB | 67.001 MiB | 80.540 MiB | 94.275 MiB | 107.843 MiB | 126.624 MiB |
| 3.100 | MinGW plain | 53.979 MiB | 67.001 MiB | 80.540 MiB | 94.275 MiB | 107.843 MiB | 126.624 MiB |
| 3.101b3 cVBR | MinGW plain | 53.006 MiB | 65.866 MiB | 79.441 MiB | 93.050 MiB | 106.432 MiB | 125.247 MiB |
| 3.101b3 cVBR Optimized | Visual Studio 2026 | 53.008 MiB | 65.867 MiB | 79.441 MiB | 93.051 MiB | 106.432 MiB | 125.249 MiB |
| 3.101b3 cVBRb Optimized | Visual Studio 2026 | 53.008 MiB | 65.867 MiB | 79.441 MiB | 93.051 MiB | 106.432 MiB | 125.249 MiB |
Optimized 3.101b3 builds versus 3.99.5
The best optimized build is approximately 11.4% faster than 3.99.5 on average across the tested ABR targets.
| Target | 3.99.5 time | Best optimized build | Optimized time | Speed gain | Output-size difference |
|---|---|---|---|---|---|
128 kbps |
39.86 s | 3.101b3 cVBRb Optimized | 34.93 s | +14.1% | -1.799% |
160 kbps |
40.01 s | 3.101b3 cVBRb Optimized | 35.80 s | +11.8% | -1.693% |
192 kbps |
38.33 s | 3.101b3 cVBRb Optimized | 34.96 s | +9.6% | -1.365% |
224 kbps |
39.34 s | 3.101b3 cVBRb Optimized | 35.66 s | +10.3% | -1.298% |
256 kbps |
40.54 s | 3.101b3 cVBRb Optimized | 36.75 s | +10.3% | -1.308% |
310 kbps |
40.98 s | 3.101b3 cVBRb Optimized | 36.56 s | +12.1% | -1.086% |
Plain 3.100 versus 3.99.5 under ABR
As in the VBR performance study, 3.100 matches 3.99.5 byte-for-byte in this ABR dataset but takes substantially longer to encode.
| Target | 3.99.5 time | 3.100 time | 3.100 slower by | Output size |
|---|---|---|---|---|
128 kbps |
39.86 s | 73.53 s | +84.5% | matches 3.99.5 |
160 kbps |
40.01 s | 75.37 s | +88.4% | matches 3.99.5 |
192 kbps |
38.33 s | 73.34 s | +91.3% | matches 3.99.5 |
224 kbps |
39.34 s | 73.16 s | +86.0% | matches 3.99.5 |
256 kbps |
40.54 s | 74.57 s | +83.9% | matches 3.99.5 |
310 kbps |
40.98 s | 75.24 s | +83.6% | matches 3.99.5 |
How this complements the VBR study
1. ABR is target-bitrate driven
Unlike -V VBR, ABR is guided by a requested average bitrate. That makes it a useful independent performance check.
2. The regression remains visible
3.100 still matches 3.99.5 output while taking much longer, showing that the slowdown is not confined to the VBR quality curve.
3. Optimized builds remain the throughput target
The optimized 3.101b3 cVBRb build wins every ABR speed test, reinforcing the SIMD/OpenMP performance work.
4. ABR has a different use case
It is most interesting for bitrate-budgeted audio: older AVI workflows, movie soundtracks, speech and audiobook-style material.
External context
LAME describes ABR as average-bitrate encoding, designed to target a chosen bitrate while retaining more flexibility than strict CBR. Hydrogenaudio similarly describes ABR as a variable-bitrate MP3 mode that targets a specified average bitrate.