About Advanced Audio Coding

AAC improves on the MP3 Perceptual Coding design to achieve higher compression ratios and better playback quality.

While the MP3 audio codec published in the MPEG-1 and MPEG-2 standards has been very successful, subsequent research explored how to reduce the bitrate and deliver better quality. Fraunhofer IIS were deeply involved again, this time with a larger cohort of collaborators.

MPEG-2 part 7 introduces the Advanced Audio Coding (AAC) standard, which supersedes MP3. Coding algorithms have been improved and new tools have been introduced to achieve a better compression ratio. It can also be used at a higher bitrate for better quality – but it is still a lossy codec.

MPEG-4 part 3 further enhances the AAC coding and adds support for other kinds of audio coding. Some of these are designed for very low bitrate transports.

Improvements Over MP3

The AAC codec improves on MP3 in several important areas:

• Compression algorithms.
• Sample rates.
• Multi-channel support.
• Base and enhancement layer encoding.
• Combining different encoders in MPEG-4.
• Additional profiles.

Compression Algorithms

MP3 provided two different but related kinds of compression:

• DCT - The Discrete Cosine Transform applied to a single frequency band.
• MDCT - The Modified Discrete Cosine Transformation applied to overlapping groups of frequency bands.

AAC is implemented purely with MDCT which significantly improves compression efficiency but means it is not backwards compatible with MP3. The number of frequency sub-bands is increased to 1024. Calculating the energy level for masking threshold control is now much more accurate and fine-grained. The basic DCT algorithm is explored more deeply in Chapter 3-2.

Sample Rates

Additional sample rates from 8kHz up to 96kHz extend the 16kHz to 48kHz range previously supported by MP3. Very low sample rates improve coding latency and are suitable for speech and telephony applications but not music.

Multi-Channel Support

Where MP3 supports only six channels of audio, AAC supports many more:

• 48 full-audio channels.
• 16 channels of low frequency effects below 120Hz.
• 16 dialog channels.
• 16 data streams.

This facilitates the coding of more sophisticated surround-sound systems.

Base & Enhancement Layered Encoding

New concepts introduce the idea of base layer encoding with optional enhancement layers using other codecs. For example, CELP (speech coding) can be improved by adding an AAC enhancement layer.

Combining Different Encoders

The General Audio (GA) coding environment described in MPEG-4 Part 3 adds two new encoders which can be used interchangeably with AAC:

• TwinVQ - Suitable for very low bitrates.
• BSAC Encoder - A scalable bitrate encoder with an error resilient bitstream.

The coding algorithms are particularly well described in the ISO 14496 Part 3 standard. Refer to sub-part 4 (around page 487) for a break down and description of each tool. The block diagram showing how these tools work in the encoder is especially helpful.

Transport Bitrates

There are several kinds of delivery for compressed audio when it is transported over a network:

Profile Support

AAC is a modular codec with a variety of tools that can be optionally switched on when needed. MPEG-2 part 7 defines several profiles which configure the earlier AAC encoder. These are inherited by MPEG-4:

• Low-Complexity Audio.
• Main Audio.
• Scalable Sampling Rate.

MPEG-4 introduces new profiles to address a wider range of applications. The coding tools configured by each profile are now redesigned and described as Audio Objects. Some profiles combine other codecs with AAC using the layered support and a few do not use AAC at all:

• Main (updated).
• AAC profile.
• Long Term Prediction.
• Scalable Audio.
• Speech Audio.
• Synthetic audio.
• High quality audio.
• Low delay audio.
• Low Delay AAC.
• Low Delay AAC V2.
• Mobile Audio Inter Networking.
• Natural audio.
• High Definition AAC.
• ALS Simple.

The entire menagerie of 42 audio objects (previously called tools) are described in sub-section 1.5.1.2 of the standard. Study these objects to better understand the profiles. Audio objects are mapped to the profiles in Table 1.3 within the standard.

Storing AAC Content In Files

MPEG standards describe these alternatives for storing AAC coded audio content in file containers:

Other containers such as QuickTime and Matroška can also be used.

AAC Implementations

Coding tools are often presented via a Graphical User Interface (GUI) wrapper for easier access. It may not be obvious at first, but the encoders are also accessible from the shell scripting environment. Command-line tools integrate more easily with workflow automation than graphical user interface applications.

Note that ffmpeg is both a command-line tool (ffmpeg) and an open-source project (ffmpeg).

The Apple AAC encoder is considered to be the best implementation for medium bitrate scenarios. The CBR, VBR and ABR support is exceedingly good. It was originally part of the QuickTime media framework but has now been moved into the AV Foundation and AudioToolbox. This is all part of what Apple calls CoreAudio. For many workflow automation situations, a macOS based processing node running this encoder will be an optimal solution.

Fraunhofer codecs are very versatile and technically the best. This is because Fraunhofer was one of the key developers of the psychoacoustic approach to audio coding. Deploying the Fraunhofer FDK AAC encoder on a Linux platform would be a very good solution but be careful to investigate and pay the licensing fees if necessary.

The Nero encoder is also highly recommended but is not being actively developed any further. It has not been revised since 2010. Whilst it is good for niche situations it is not recommended for new project deployments.

The rest of the AAC implementations are somewhat lacking in their support for all the different modes of operation. There are a few implementations based on the libraries developed by Coding Technologies. They collaborated with Fraunhofer on the AAC research. These may incur license fees when deployed.

Deploying The Apple AAC Encoder In A Workflow

Deploy a processing node in your workflow based on a single Macintosh computer to call the Apple AAC encoder to action from the command-line.

Run a frequently scheduled task to check a watch folder for input files. This is easy to implement. The macOS environment supports all the tools you need to pass the output to the next stage of the workflow.

File Name Extensions

These file extensions are relevant when coding MPEG AAC Audio:

Media Type Identifiers

Relevant Standards

Consult these standards documents for background information:

Applying AAC

AAC coding will be subject to patent licensing fees until 2031. For the time being you will need to contact a patent pool for a license if you build and distribute an encoder or player implemented in hardware or software. The coded bitstreams transmitted to end users are free of any licensing obligations.

This is not the whole AAC story. High Efficiency AAC was developed to improve the performance still further. That is sometimes called AAC+, and we look at this in more detail in the next chapter.

There is also more to study and understand in the MPEG-4 (ISO 14496) standard. MPEG is reorganizing the collection of standards and MPEG-D (ISO 23003) has some relevant material and ISO 23091-3 is helpful for player design.

These Appendix articles contain additional information you may find useful: