Encoding throughput
507× realtime
3.3× the measured FDK-AAC throughput
FAAC506.9×
FDK-AAC152.9×
Apple AAC112.1×
FFmpeg AAC40.4×
Aggregate throughput · higher is better
AAC Benchmarks: Speed, Size, and Quality
Compare measured encoding speed, code footprint, low-bitrate quality, and decoder performance. The results below are a published benchmark snapshot; the methods explain how these measurements guide FAAC tuning.
FAAC encoder speed and footprint
Encoder footprint
73 KB
About 8% of the measured FDK-AAC footprint
FAAC73.4 KB
FDK-AAC940.5 KB
Apple AAC100.3 KB
FFmpeg AAC274.5 KB
Code + constants + initialized data · lower is better
Benchmark snapshot · September 29, 2026 · MacBook Pro M1, 32 GB. The aggregate report covers 27 scenarios. Tested builds: FAAC 2.2.0 (6acbe23a), FDK-AAC 2.0.3 through fdkaac 1.0.9, Apple AAC 27.0, and FFmpeg g3fc4f99a79 with experimental NMR. Published measurements.
Low-bitrate quality: FAAC vs FFmpeg
See FAAC's HE-AAC v1 results against FFmpeg's native LC encoder, or select LC vs LC for a same-profile comparison.
Average objective MOS · higher is better
24 kbps
FAAC · HE v13.43
FFmpeg · LC2.52
32 kbps
FAAC · HE v13.81
FFmpeg · LC2.51
40 kbps
FAAC · HE v14.02
FFmpeg · LC2.67
48 kbps
FAAC · HE v14.17
FFmpeg · LC3.83
56 kbps
FAAC · HE v14.31
FFmpeg · LC4.13
64 kbps
FAAC · HE v14.41
FFmpeg · LC4.28
Different profiles: FAAC uses HE-AAC v1; the tested FFmpeg native encoder uses AAC-LC. Bitrates are requested targets, not matched actual payload rates. Scores are model estimates, not listening-test results.
View all scores
| Target kbps | FAAC LC | FAAC HE v1 | FFmpeg LC |
|---|---|---|---|
| 24 | 2.715 | 3.427 | 2.515 |
| 32 | 3.454 | 3.812 | 2.508 |
| 40 | 3.868 | 4.020 | 2.667 |
| 48 | 4.110 | 4.172 | 3.833 |
| 56 | 4.258 | 4.309 | 4.129 |
| 64 | 4.379 | 4.414 | 4.279 |
Average scores from the same snapshot. See the report for profile breakdowns and bitrate accuracy.
Decoder performance: FAAD vs Helix and others
Decoding throughput
490× realtime
FAAD3 · 15% higher throughput than LibHelix
FAAD3490.3×
LibHelix426.1×
FAAD2376.8×
FDK-AAC344.1×
Apple AAC174.4×
FFmpeg AAC135.1×
Aggregate throughput · higher is better
Decoder footprint
54% less code
FAAD3 compared with FAAD2 · 136 KB
FAAD3136.0 KB
LibHelix121.4 KB
FAAD2296.3 KB
FDK-AAC940.5 KB
Apple AAC100.3 KB
FFmpeg AAC274.5 KB
Helix and Apple were smaller · lower is better
The same published report compares released FAAD2 and the FAAD3 development engine with other AAC decoders.
| Decoder | Throughput ↑ | Code footprint ↓ | Peak RAM ↓ |
|---|---|---|---|
| FAAD2 2.11.3 | 376.8× | 296.3 KB | 2.1 MB |
FAAD3 (f9fbcb76) | 490.3× | 136.0 KB | 2.8 MB |
| Helix AAC 1.0 (LibHelix) | 426.1× | 121.4 KB | 1.9 MB |
| FDK AAC 2.0.0 | 344.1× | 940.5 KB | 4.2 MB |
FFmpeg AAC (g3fc4f99a79) | 135.1× | 274.5 KB | 16.8 MB |
| Apple AAC 27.0 | 174.4× | 100.3 KB | 8.3 MB |
The tradeoff: FAAD3 was faster and used less code storage than FAAD2, with higher peak RAM. Helix was faster and smaller than FAAD2; FAAD3 was faster than Helix, with a larger footprint.
Helix lacks HE-AAC v2 Parametric Stereo (PS) support. For FPU versus fixed-point hardware guidance and release requirements, see choosing a decoder.
How audio quality is tuned
The FAAC benchmark suite evaluates mono speech, stereo music, and multichannel audio across sample rates and bitrates. Fine-tuning considers difficult passages alongside average quality.
- Speech and music quality: ViSQOL and Zimtohrli estimate changes relative to the source.
- Difficult passages: lower-tail scores help expose regressions hidden by an average.
- Sharp attacks: onset timing measurements track changes around percussion and plucked strings.
- Stereo image: coherence measurements track changes between the left and right channels.
- Frequency balance: spectral diagnostics locate regions where distortion changes.
These measurements guide investigation; objective scores do not establish audible preference.
How bitrate efficiency is checked
Quality comparisons use the actual AAC payload bitrate, excluding M4A headers and metadata. This prevents extra bits or container overhead from disguising a rate-control difference.
Rate–quality curves and BD-rate analysis compare the bits needed to reach the same objective quality. Sample-rate and channel groups stay separate, as do AAC-LC and HE-AAC v1. ABR, VBR, and CBR are evaluated as distinct modes.
How speed and footprint are checked
Benchmarking tracks three resource costs:
- Throughput: audio duration encoded per unit of processing time.
- Code footprint: executable code, constant tables, and initialized data.
- Memory: writable static data and peak process memory.
Baseline and candidate builds record their toolchain and settings. Parameter sweeps, decode validation, and regression reports help weigh a tuning change's quality gain against its processing and storage cost.
For your application, measure the selected build on the target device. Encoding throughput alone does not establish live-audio delay or power consumption.
Benchmark methodology
The benchmark documentation defines the quality and efficiency metrics, test scenarios, and footprint checks.
Next: choose an encoder or decoder, or test audible quality.