What is AAC? Advanced Audio Coding Explained

AAC (Advanced Audio Coding) is a lossy audio codec standardized by MPEG in 1997 as the successor to MP3, and it remains the default audio format for video on the internet. It delivers the same perceived quality as MP3 at roughly 70% of the bitrate, supports up to 48 channels and sample rates from 8 to 96 kHz, and is decodable by virtually every phone, browser, TV, and computer in use today. If a video plays sound on the web, there is a good chance that sound is AAC.

For Videospan, AAC is the finish line of every video we produce. When a brand video leaves our platform — as a download, a repurposed clip, or a stream — its audio is AAC, because it is the one codec we can guarantee every viewer's device will decode. This guide covers how AAC works, why it earned that position, and where it sits in our stack.

How AAC works

Like every modern perceptual codec, AAC is built on a psychoacoustic model: it analyzes the audio to determine what listeners can actually hear, then spends its bitrate budget only on that.

The pipeline works like this:

1. Frequency analysis. The signal is transformed from the time domain into the frequency domain using a Modified Discrete Cosine Transform (MDCT). AAC switches between long blocks (2048 samples) for steady, tonal content and short blocks (256 samples) for transients like drum hits, so sharp attacks don't smear across time.
2. Masking thresholds. A psychoacoustic model computes, per frequency region, the masking threshold — the level below which quantization noise will be inaudible because louder nearby sounds (in frequency or time) hide it.
3. Quantization by scalefactor bands. The MDCT coefficients are grouped into ~49 scalefactor bands that approximate the ear's critical bands. Each band gets just enough bits to keep coding noise under its masking threshold; content below the threshold gets few bits or none.
4. Entropy coding. The quantized coefficients are Huffman-coded and packed, along with the scalefactors, into the bitstream. Optional tools like Temporal Noise Shaping (TNS) and Perceptual Noise Substitution (PNS) handle difficult signals more efficiently.

The result: a 128 kbps stereo AAC stream keeps everything above the masking thresholds and discards the rest — which is why it can sound indistinguishable from the 1.4 Mbps original in blind listening tests.

AAC profiles: LC, HE-AAC, and HE-AAC v2

AAC is a family of profiles, not a single format:

• AAC-LC (Low Complexity) is the workhorse. It offers the best quality per bit at normal bitrates (96 kbps and up) and is required by every baseline decoder. This is what almost all video uses.
• HE-AAC v1 (High Efficiency) adds Spectral Band Replication (SBR): the codec encodes only the lower frequencies and transmits a compact description that lets the decoder reconstruct the high frequencies. It shines at low bitrates (32–64 kbps) where AAC-LC starts to sound muffled.
• HE-AAC v2 adds Parametric Stereo on top of SBR: at very low bitrates (24–48 kbps), stereo is encoded as a mono downmix plus a small set of spatial parameters. It is common in low-bandwidth streaming and broadcast.

A practical rule: use AAC-LC for anything at or above 64 kbps, and consider HE-AAC when bandwidth is genuinely constrained.

Seeing the bandwidth cutoff

Perceptual coding has a visible signature in the frequency domain. At 128 kbps stereo, AAC typically retains content up to about 18–20 kHz and discards the highest octave entirely; at lower bitrates the cutoff moves down. The demo below shows a real spectrum with the AAC cutoff line — everything above it is information the psychoacoustic model decided you can't hear.

How Videospan uses AAC

AAC is our delivery standard: every video our platform renders — a full-length executive recording, an AI-assisted edit, a repurposed short clip in any aspect ratio — leaves with an AAC audio track, so the finished file plays on every phone, browser, and TV without fallback logic. The same holds for streaming output: HLS mandates AAC support across the Apple ecosystem, which makes it the only audio codec we can guarantee everywhere.

Speech clarity is where this matters in practice. Most corporate video is talk: an executive recorded on a laptop microphone still needs to sound crisp after encoding. AAC-LC keeps the vocal range intact at 96–128 kbps stereo, and speech-only content stays clean at 64 kbps — or 48 kbps with HE-AAC — so exports stay high-quality without bloating file sizes or delivery costs.

That universality is also what makes AAC the right last step in a browser-based capture pipeline. Contributors record on whatever hardware and network they have; the audio that comes out the other side has to be unremarkable in the best way — it just plays, everywhere. That is AAC's job in our stack.

AAC vs Opus

AAC and Opus overlap in quality at normal streaming bitrates, but they were designed for different jobs.

The short version: AAC wins on compatibility and is the safe default for video delivery; Opus wins on latency, low-bitrate quality, and licensing, and is the standard for real-time communication.

Advantages

• Universal playback. Every modern OS, browser, phone, TV, and set-top box decodes AAC natively — no plugins, no fallback logic.
• Required by HLS. Apple's HLS specification mandates AAC (in fMP4 or ADTS segments) as a supported audio format, and it is the only audio codec guaranteed to work across the entire Apple ecosystem.
• Excellent quality at standard bitrates. 128 kbps stereo AAC-LC is transparent or near-transparent for most listeners and content.
• Mature tooling. Every encoder, packager, CDN, and analytics platform handles AAC without special configuration.
• Flexible profiles. From parametric stereo at 24 kbps to multichannel at 48 channels, one codec family covers the whole range.

Limitations

• High algorithmic delay. Frame sizes, the MDCT window, and the SBR look-ahead add up to roughly 100 ms or more of unavoidable delay — unacceptable for two-way real-time communication.
• Patent licensing. AAC is covered by patents administered through the Via LA pool. Encoding and distribution of encoders technically requires a license, although decoder implementations and end-user playback are royalty-free in practice, which is why it never blocks typical video workflows.
• Not optimal at very low bitrates. Below ~48 kbps, AAC-LC degrades noticeably; HE-AAC helps, but Opus generally sounds better in that range.

When to use AAC

Use AAC whenever you are delivering prerecorded audio or video to a general audience:

• Video on the web and in apps. AAC in an MP4 container plays everywhere. For stereo, 96 kbps is a solid floor and 128 kbps is the common transparent target.
• HLS streaming. AAC is effectively mandatory: it's the one audio codec every HLS client is required to handle. Multiplex it with H.264/H.265 video in fMP4 segments for maximum reach.
• Podcasts and downloads. .m4a (AAC in MP4) is the standard for audio distribution.
• Anything targeting Apple devices. Safari, iOS, AirPlay, and CarPlay are AAC-native end to end.

Encoding is a one-liner with ffmpeg:

# Video delivery: AAC-LC stereo at 128 kbps in MP4
ffmpeg -i in.wav -c:a aac -b:a 128k out.m4a

# Low-bandwidth stream: HE-AAC at 48 kbps
ffmpeg -i in.wav -c:a aac -profile:a aac_he -b:a 48k out.m4a

Reach for Opus instead when latency matters (video calls, live interactive, WebRTC) or when you must stream at bitrates below ~64 kbps.

If you're evaluating a video platform, audio delivery is worth one direct question: what codec do exported files and streams carry, and will they play on every device your audience owns without a fallback? Any answer other than AAC (or AAC plus a fallback track) means some share of your viewers gets silence — or a support ticket. Ask for an exported file and play it on the oldest device in your office.

Glossary

Psychoacoustics

The study of human sound perception. AAC's encoder uses a psychoacoustic model to decide which parts of the signal are inaudible and can be discarded.

MDCT

Modified Discrete Cosine Transform. The time-to-frequency transform at AAC's core, using overlapping blocks (2048 or 256 samples) to represent the signal for quantization.

Masking

When a loud sound renders quieter nearby sounds inaudible, in frequency or in time. AAC hides its quantization noise under masked regions.

Scalefactor bands

Groups of MDCT coefficients approximating the ear's critical bands. AAC allocates bits per band to keep coding noise below the masking threshold.

Profile

A defined subset of AAC's tools. AAC-LC is the baseline; HE-AAC adds Spectral Band Replication; HE-AAC v2 adds Parametric Stereo.

Spectral Band Replication (SBR)

An HE-AAC technique that encodes only low frequencies and transmits parameters letting the decoder reconstruct the high band, improving low-bitrate quality.

Bitrate

Bits per second of compressed audio. For stereo AAC-LC, 96–128 kbps is the normal video-delivery range.

Sample rate

Samples per second of the digital signal. AAC supports 8–96 kHz; 48 kHz is standard for video.

Algorithmic delay

Latency inherent to a codec's frame structure and analysis window. For AAC it is roughly 100 ms or more, ruling out real-time use.

Frequently asked questions

Related

• Opus: the open, low-latency audio codec
• Audio codecs overview: AAC & Opus
• MP4 container
• HLS streaming