To date, the AAC format has not yet reached mass distribution on audio media, however, in a number of parameters it surpasses all types of audio compression that exist today, which means it deserves our attention.
What it is?
Let's start with a definition: AAC is a proprietary (proprietary) compression option for an audio file. At the same time, it has less quality loss during encoding compared to MP3 at the same bitrate. In addition, the AAC format is a wideband audio coding algorithm that uses two main coding principles to greatly reduce the amount of data required to transmit quality digital audio. This solution is recognized as one of the highest quality implemented using lossy compression technology. The format is supported by most modern equipment, even portable ones. It should be noted that AAC ringtones can be purchased from the iTunes Store, and this store only offers music compressed with this solution. It must also be said that the AAC format was originally created as a successor to MP3, which can provide improved encoding quality. The solution was published back in 1997 as a new, 7th, part of the MPEG-2 family.
Principle of operation
When encoding into this format, the following processes are performed: imperceptible components are removed from the signal, the encoded audio signal is cleared of redundancy. After that, the data is processed in accordance with the MDCT method according to their complexity. At the next stage, codes are added to correct various internal errors. Finally, the signal is transmitted or stored.
All the details
Interestingly, the AAC format has a sampling rate in the range of 8-96 kHz, as well as the number of channels in the range of 1-48. MP3 uses a hybrid set of filters. In turn, AAC refers to the Modified Discrete Cosine Transform with an increased window size, which reaches 2048 points.
Thus, AAC is much more suitable for encoding audio having complex pulse stream as well as square wave as compared to MP3. The format has gained dynamic switching capability in MDCT block lengths in the range of 2048-256 points. In case there is a short or single change, a small window of 256 points is applied in order to achieve a better resolution. This defaults to a 2048-point large window to maximize encoding efficiency. AAC has a number of advantages over conventional MP3. Among them, it should be noted: the implementation of a large number of audio channels (up to 48), significant coding efficiency in conditions of constant and variable bitrate, as well as sampling rates ranging from 8 Hz to 96 kHz (for MP3 this figure is from 8 Hz to 48 kHz) and a more flexible special mode called Joint stereo. As for the solution, "AAC +" is a codec that is focused on working with a low bitrate. It is a combination of SBR and AAC LC, thanks to which good sound is already achieved in the range of 32-48 kbps.
File extension: .m4a , .aac , .mp4
Specifications for the most common AAC LC mode:
The format is part of the MPEG Audio standard. ISO/IEC 13818-7. It was created by the German institute Fraunhofer IIS and is a further development of the MP3 format.
The format itself is proprietary and requires the purchase of a license for commercial use.
Story
The development of the format began in 1994 by the joint efforts of Fraunhofer IIS, AT&T, Dolby and Sony. It wasn't until three years later that the format became part of the MPEG standard as MPEG-2 AAC. With the subsequent development of the MPEG-4 audio standard, AAC has been improved and refined.
In general terms, the chronology of the development of the format is as follows:
- 1997 - standardization MPEG-2 AAC-LC.
- 1999 - standardization MPEG-4 AAC-LC. Added PNS (Perceptual Noise Substitution) technology.
- 2003 - standardization MPEG-4 HE-AAC. Added SBR (Spectral Band Replication) technology.
- 2004 - standardization MPEG-4 HE-AAC v2. Added PS (Parametric Stereo) technology.
HE-AAC is a low bitrate oriented format. The combination of AAC LC and SBR used in it gives good quality at bit rates from 32 to 48 kbit/s. Naturally, HE-AAC supports multi-channel and allows for a rich selection of sample rates. HE-AAC also known as aac Plus.
When combined with parametric stereo HE-AAC v2 provides good audio quality at bitrates of around 16 kbit/s for stereo. HE-AAC v2 also known as aac Plus v2.
Profiles
- LTP / Long Term Prediction, MPEG-4 only
Some profiles are of low complexity and are therefore preferred for portable devices, but they also provide slightly lower quality. However, the most common profile currently is , as the other more demanding profiles (Main , LTP) do not provide as much quality as they are more computationally demanding.
Titles
There are 4 types of headers in AAC:
- LATM (Low-overhead MPEG-4 Audio Transport Multiplex)
- LOAS (Low overhead audio stream)
ADIF
This header format is intended for simple local storage, unlike ADTS and LATM/LOAS which are intended for AAC mobile transmission.
ADTS
qaac
There is an open source wrapper implementation for AAC from Core Audio and ALAC called qaac by developer nu774 that only works under Windows. At the time of updating the article, this is the most current AAC codec.
Nero Digital Audio
Starting with the fifth version, the Nero package comes with the AAC codec. Now he is part of Nero Digital - a project to create a new standard in the world of digital video and audio. Nero AAC can create two types of AAC: LC and HE. Low Complexity (LC) is the usual AAC standard, while High Efficiency (HE) uses SBR technology (similar to mp3PRO), and if the player understands this technology, additional frequency information can be decoded, otherwise half of the audio quality is lost during playback. At the moment, the codec is significantly outdated and its quality is inferior to the implementation from Apple. This codec is very convenient to use together with dBpowerAMP Music Converter and its corresponding codec (dBpowerAMP Nero MP4 Codec).
Advantages of Nero MPEG-4 AAC (LC profile):
- In all cases, the quality is noticeably better than LAME MP3 at smaller file sizes.
- Low resource consumption compared to the latest versions of LAME MP3 (VBR/ABR).
Psytel AACenc and Fastenc
PsyTEL® FAST MPEG-2 AAC LC Encoder v2.0 (build Mar 4 2002) / Copyright © 1999-2001 PsyTEL Research / Copyright © 1999-2001 Ivan Dimkovich
The Yugoslav company Psytel Research has been working on improving the MPEG-2/4 AAC standard. It was founded around 1998 by programmer Ivan Dimkovich.
While the first version (1.0) was roughly a compilation of ISO source codes, the quality began to increase rapidly and, after a while, it was the only truly ISO optimized AAC encoder available to the public (the FhG and Dolby versions were only available to developers interested in in improving their technology). The Liquifier was also available, but the data streams it created were encrypted.
By mid-2002, Psytel Research and all of its assets were bought by Ahead Software GmbH., and AACEnc became Nero AAC Encoder.
FAAC (Freeware Advanced Audio Coder)
At the end of 1999, Menno Bakker opened the source codes of his developments and justified the FAAC (Freeware Advanced Audio Coder) project. The FAAD2 decoder is considered one of the best and fastest AAC decoders today. The FAAD2 (Freeware Advanced Audio Decoder) project has Copyright © 2003-2004 M. Bakker, Nero Software AG. For licensing, please contact Nero Software AG.
HomeBoy AAC
HomeBoy was the name of a group of programmers who created the first ISO publicly available AAC encoder for Windows back in 1998. They were also reportedly the creators of the first third party plugin for Winamp (their AAC input plugin), thus making it available to the general public. first ISO AAC decoder. The encoder is just a compilation of the original ISO-linked sources, so the quality was poor. But what is interesting is that the data streams created by him are still decoded in modern decoders.
Versions
Dicas/zPlane Compact!
AAC Codec Compaact! was developed by German DSP engineer Alexander Lerch. The release took place in 2003. The codec was well received upon release. It also contained some interesting features. There were three levels of quality (from fast to high quality encoding), supported Main profiles, it was possible to enable or disable technologies such as TNS or PNS, support for multi-channel encoding, some pre-processing operations and a very interesting pre-listening feature that allowed you to hear in real time how the songs will sound after compression. For unknown reasons, the project was stopped in 2005.
Versions
MBsoft AAC
The project started around 1998 when German programmer Menno Bakker started his work on the AAC source codes. encoder mbaacenc was more or less an ISO source build with a good front-end. A plugin for Winamp was also available.
At the end of 1999, Menno opened the source codes of his developments and justified the FAAC (Freeware Advanced Audio Coder) project. The project is still alive and its decoder, FAAD2, is considered one of the best and fastest AAC decoders today.
I would like to introduce you to the same audio format as AAC.
What are the advantages of this AAC format over others, for example over mp3?
Up to 48 audio channels;
Greater coding efficiency at both constant and variable bitrates;
Sampling rates from 8 Hz to 96 kHz (MP3: 8 Hz - 48 kHz);
More flexible Joint stereo mode.
> Simply put, AAC is not only a better format than others, but it also has advantages.
Do you still store music in MP3 and LOSSLESS?
Our answer to that is AAC only! Judge for yourself: the largest social networks use this format in their videos, which saves a lot of space on their hard drives. The format is supported by most modern phones that can play mp3, wma. For example, Nokia Corporation records the standard ringtones that are in their phones in the AAC format. Are you not yet convinced of its superiority? Then let's move on to the numbers...
When compressed from MP3 to 3GPP AAC+ format, the music file loses its weight by at least THREE times!
That is, we take a file with a sound bitrate of 320 kbps, and at the output we get a bitrate of 48 kbps, We consider 320/48=6.666. That is, the MP3 file can potentially be reduced to SIX TIMES!
Now let's imagine that your 40GB audio library in MP3 format is reduced by a factor of 40/3=13! TOTAL THIRTEEN GB! However, the sound quality WILL NOT HAVE ANYTHING. If you are in doubt, I will say this: I store all my music in AAC format, so three years ago I had a 40GB hard drive. Yes, yes, don't be surprised) So I started looking for methods to reduce my audio library. I didn't want to delete the music, I just converted everything to AAC. Of course, on my AMD 1500+ this happened in stages and not as fast as I would like, but I did it!
Want more information?
To date, manufacturers of hard drives are interested in what their product would be bought. In this regard (believe it or not) you are forced to download more and more, store the most information on your disks, in the most cumbersome file size. For example, Modern BluRay video formats. I do not argue, if you have a giant screen, then you just have to use this format, but if you use ordinary monitors, up to 22", then tell me why you should watch movies in a format whose picture image is larger than your monitor?
What about LOSSLESS?
Guys, finally understand that this format was created for special use in those places where it is really needed. For normal listening through the player on the computer HE'S NOT NEEDED! Lossless is be h useful waste of disk space. By ear, a person cannot perceive the difference between high-quality AAC and Lossless. Therefore, I recommend this format to you. Of course, you will immediately react with disbelief, but ... I would advise you to try it. And you will understand for yourself that you can not find the best!
What are the disadvantages of AAC?
I understand the advantages of this format, but how can I transcode my music?
I would advise you to use mediacoder. This is a great encoder for your media files. Supports a huge number of formats, is h paid.
How exactly to encode?
2. Install in a couple of clicks, click Next>...
3. Open the folder with your music, drop it into mediacoder and choose a format.
2009-09-30T20:52
2009-09-30T20:52
Audiophile's Software
The first ideas about using psychoacoustic masking to compress audio data date back to 1979. However, the corresponding audio encoders began to become widespread only from the mid-90s, when the computing power of personal computers began to be enough to play compressed audio in real time and the MPEG-1 Audio Layer 3 standard, better known as MP3, appeared. Compressed audio formats have become indispensable for transmitting audio over the Internet, providing "virtually transparent" stereo sound quality (that is, the encoded signal is indistinguishable from the original for most listeners) at bit rates above 128 kbps. The basic principles of the MP3 format can be found in the articles by K. Glasman (2…8/2005)
The development of data compression methods and psychoacoustics gradually led to the fact that the MP3 standard became "close" to the implementation of new ideas in audio coding. As a result, by 1997, the Fraunhofer Institute (Fraunhofer IIS), which created MP3 in the early 90s, as well as Dolby, AT & T, Sony and Nokia, developed a new audio compression method - Advanced Audio Coding (AAC), which was included in the standards MPEG-2 and MPEG-4. The main differences from the MP3 standard are:
- support for a wider range of formats (up to 48 channels) and audio sampling rates (from 8 kHz to 96 kHz);
- more efficient and simpler filter bank: MP3 hybrid filter bank has been replaced by conventional MDCT (Modified Discrete Cosine Transform);
- wider limits of variation in the frequency-time resolution in the filter bank - eight times (in MP3 - three times) - led to improved coding of transients (transients) and stationary sections of the audio signal;
- better coding of frequencies above 16 kHz;
- a more flexible stereo coding mode that allows you to switch to M / S (“joint stereo”) mode independently in different frequency bands;
- additional features of the standard that increase the efficiency of compression: technology for generating noise in the time domain (TNS), prediction of MDCT coefficients in time (long term prediction), parametric stereo coding mode (parametric stereo), noise synthesis (perceptual noise substitution), high-frequency frequencies (SBR).
Thanks to these features, the AAC standard is able to achieve more flexible and efficient, and thus better quality audio coding. As a result of the widespread use of the MP3 format, the AAC standard has not yet acquired the popularity comparable to MP3. However, AAC is the main format in the popular iTunes Store, iPod, iTunes, iPhone, PlayStation 3, Nintendo Wii and DAB+/DRM digital broadcasting.
Consider the main features of AAC in more detail.
Filter bank
Like other psychoacoustic audio encoders, AAC works in the following way. The input signal is passed through a filter bank - a transformation that converts the signal from the time domain to the frequency-time domain (similar to building a spectrogram). In parallel with this, the psychoacoustic model analyzes the signal and determines the thresholds for psychoacoustic masking. Next, the spectral coefficients of the signal at the output of the filter bank are quantized so that the noise spectrum, if possible (if the bit rate permits), is below the masking thresholds and is not audible. The quantized coefficients are compressed losslessly into an AAC output file. Thus, the filter bank itself does not compress the signal, it only converts it into a form more suitable for compression.
A feature of each filter bank is its frequency resolution, that is, the number of frequency bands into which it divides the signal spectrum. Most filter banks used for audio compression have hundreds of bands. This means that, due to the uncertainty relation, such filter banks have a time resolution of the order of several tens of milliseconds. When the spectral coefficients of a signal are quantized, the introduced quantization error during signal decoding is spread over time over the entire length of the filterbank window. In some cases, this leads to an unwanted effect called pre-echo. It manifests itself when the quantization error from the transient (a sharp burst of energy in the signal) propagates in time to the time segment preceding the transient and becomes audible (Fig. 1). To reduce this effect, filter banks with variable time-frequency resolution are used. For example, MP3 uses a switchable filter bank time resolution between 26 and 9 ms. Stationary signals use 26 ms windows to give good frequency resolution, while transients use 9 ms windows to reduce the pre-echo effect (see Figure 1).
The AAC algorithm also uses MDCT window size switching. In this case, the difference in the size of the windows is eightfold: 6 and 48 ms (256 and 2048 samples). Due to this, the algorithm is able to adapt to a wider range of signals and achieve a better degree of compression.
TNS Technology - Noise Amplitude Envelope Shaping
One of the problems of modern psychoacoustic audio signal encoders is working with transients (transients in an audio signal). To ensure transparent coding, it is necessary to ensure that the quantization noise falls under the masking threshold, which depends on time. However, in practice, this requirement is difficult to satisfy near transients, because The quantization noise generated during encoding is spread in time during decoding over the entire length of the MDCT window. This can lead to significant excesses of quantization noise over time masking thresholds.
TNS (temporal noise shaping) technology in the AAC standard allows you to control the propagation of time quantization noise within each MDCT window. The TNS technology is based on the similarity (time-frequency dualism) of the amplitude envelope of the signal and the envelope of its spectrum, as well as the use of linear prediction (LPC) in frequency when quantizing the spectrum.
It is well known that for signals with a spectrum that is very different from white (for example, tones), the use of linear prediction (LPC) in the time domain allows you to effectively "whiten" the spectrum and encode such signals by decomposing them into prediction coefficients and a relatively small amplitude prediction error (residual). When decoding, the linear prediction filter generates an error spectrum according to the spectrum of the original signal.
In the AAC encoder, linear prediction is used in the opposite way: to predict spectrum samples in the frequency domain. The difference between the original and predicted MDCT coefficients is quantized according to concealment thresholds (in conventional encoders, the original MDCT coefficients are quantized). Linear prediction coefficients are also written to the output file. When decoding a signal, a linear prediction filter applied to a difference signal in the frequency domain (including quantization error) generates an amplitude envelope of the original signal (and quantization error) in the time domain. Thus, the amplitude envelope of quantization errors becomes close to the amplitude envelope of the original signal (Fig. 2).
TNS technology reduces the effect of pre-echo and the visibility of quantization errors on some pulsed harmonic signals (speech, some wind and bowed string instruments). On fig. 2 compares the quantization errors introduced into the vocal signal by the AAC and MP3 algorithms with the same bitrates. Together with a general decrease in the quantization error (due to the greater efficiency of AAC), the formation of the amplitude envelope of the quantization error in time is observed according to the envelope of the original signal.
In the AAC standard, TNS technology can be applied to individual frequency bands of the spectrum independently or turned off completely.
SBR Technology - Treble Restoration
Reliable transmission of a wide frequency range is an important requirement for high-quality coding. However, the transmission of each next octave of the audio range increases the bitrate requirements for a traditional audio encoder by one and a half to two times. In order to reduce the bitrate and at the same time preserve high frequencies in the encoded material, a technology for artificially synthesizing high frequencies SBR (spectral band replication) was created.
The technology is based on the fact that our hearing analyzes high frequencies with less accuracy than medium and low ones. To create the effect of the presence of high frequencies, it is not necessary to reconstruct the waveform mathematically exactly, but it is enough just to restore some significant psychoacoustic signal parameters at high frequencies. These essential parameters include the time-frequency distribution (envelope) of signal energy and the degree of its tonality/noisiness.
The idea of the algorithm is this. When encoding, the analysis of high frequencies in the original audio signal is carried out and their parameters are extracted: first of all, the amplitude envelope in several (usually eight) frequency bands. Further, high frequencies are removed from the record and only the remaining low and medium frequencies are encoded. In this case, a relatively small stream of information about the parameters of the lost high frequencies is also added to the output file.
During playback, the bass and mid frequencies are decoded first. Further (if it is present in the player), the SBR decoder starts working. The first step is to synthesize a high-frequency signal by transposing (more precisely, a frequency shift) the available mid frequencies. Since the degree of tonality/noisiness of the spectrum at medium and high frequencies is approximately equal, as a result of this step, a high-frequency signal with a plausible spectrum structure is obtained. In the second step, the SBR decoder uses the additional stored information about the high frequencies to give them the desired amplitude envelope in each frequency band. The result is a signal whose high frequencies are fully synthesized from the mids, while retaining the sound of the original high frequencies.
SBR technology can be added to many existing audio coding methods. For example, SBR combined with MP3 is called MP3 PRO, and SBR combined with AAC is called HE-AAC (high efficiency AAC). Basically, SBR is used for encoding at relatively low bit rates: 64 kbps and below. The technology allows you to significantly expand the frequency range of the audio signal with a minimal increase in bitrate (several kbps).
Parametric stereo technology
The transfer of a stereo signal usually requires almost 2 times more bitrate from the encoder than the transfer of a mono signal. At the same time, stereo channels can be encoded both independently and after M/S conversion. In the latter case, the S-channel often consumes less bitrate than the M-channel. This encoding mode is also called joint stereo. In the AAC standard, this mode can be turned on and off by the encoder independently for each frequency band.
For more efficient coding of stereo signals at very low bitrates (16...32 kbit/s), the technology of parametric stereo coding (parametric stereo) was developed. It lies in the fact that the stereo signal is reduced to mono before encoding, but a small stream (2 ... 3 kbit / s) is added to the output file, containing information about the stereo panorama of the original stereo file. This stream contains (in a compressed form) a kind of "panorama map" for the time-frequency plane.
In the decoding stage, frequency-dependent panning is applied to the received mono signal. This can be done simultaneously with decoding by applying appropriate amplitude factors to the initially equal MDCT coefficients of the left and right channels.
Parametric stereo technology gives a good impression of the original stereo sound at the cost of only a small increase in bitrate compared to mono encoding. However, it does not allow you to achieve completely transparent sound, as it is unable to take into account all the nuances of the stereo panorama, such as phase shifts between stereo channels.
Parametric stereo technology has been included in the HE-AAC v2 standard.
PNS technology - noise generation
To further increase the efficiency of coding noise signals, the AAC standard provides for PNS (perceptual noise substitution) technology for noise synthesis. It is known that our ear is more sensitive to the amplitude spectrum of the signal than to the phase spectrum. Therefore, instead of encoding the MDCT coefficients of the original signal in the noise domains, one can only transmit noise parameters: its power as a function of frequency and time.
This is how PNS technology works. During coding, spectral regions that are noise are identified, and the corresponding groups of MDCT coefficients are excluded from the coding process. The frequency band is marked as noisy and the total noise energy is stored for it.
During decoding, pseudo-random MDCT coefficients with the required total power are substituted into the frequency bands marked as noise. As a result, in the specified frequency ranges, noise is synthesized that is close in sound to the original noise.
Long term prediction technology - time prediction
Psychoacoustic tone coding requires a higher local signal-to-noise ratio than noise coding (eg, 20 dB and 6 dB, respectively). And this, in turn, requires an increased bitrate. However, the MDCT tone coefficients are predictable over time. This circumstance allows us to exploit their time dependence to reduce the bitrate.
The AAC standard provides a Long term prediction mode in which the MDCT coefficients are additionally coded in time using linear prediction. The term "long term" means that the prediction is carried out not by neighboring samples, but by samples separated by the most probable tone period at a given frequency.
Quantization and compression of MDCT coefficients
Similar to the MP3 standard, AAC uses non-linear quantization of MDCT coefficients and their compression by the Huffman method. The MDCT coefficients are quantized after exponentiation of 0.75, which allows increasing the quantization error for strong signals and decreasing it for weak signals within each frequency band. Thus, an additional implicit formation of the noise spectrum is carried out.
After quantization, the MDCT coefficients are compressed using a set of fixed Huffman tables. In the AAC standard, there are more of these tables than in MP3, and there are more opportunities for grouping coefficients. This results in an additional increase in compression.
Sound quality
When evaluating the sound quality of audio encoders, subjective tests are usually used. Listeners are presented with fragments of recordings compressed by different encoders, and they evaluate the purity of the sound of each fragment on a scale from 1 to 5. The best codec is the one that is able to achieve higher sound quality compared to competitors at a given bitrate.
A rather authoritative Internet source, where the results of such tests are given, is the site http://www.rjamorim.com/test/ It presents tests of various codecs at a variety of bitrates. These results generally agree well with other sources. Here are some results for MP3 and AAC encoders to help compare their quality.
The best MP3 encoder is the free Lame. However, at most bitrates, it is inferior in quality to newer compression standards. At high bitrates (above 128 kbps), this lag is small, and the Ogg Vorbis encoder is the leader.
At 64 kbps, the advantage of AAC is already noticeable. In the HE-AAC variant, the algorithm earns a score of 3.68. This roughly corresponds to Lame with a bitrate of 96 kbps and means the advantage of AAC over MP3 is about 1.5 times. The Lame score at 128 kbps is 4.29.
At a bitrate of 32 kbps, Nero's AAC encoder wins significantly in quality compared to MP3: scores of 3.23 and 1.72, respectively. However, AAC is only slightly ahead of the MP3PRO format, which received a score of 3.08. This indicates that the SBR technology does significantly improve quality at low bit rates.
conclusions
Thanks to the new technologies used in the AAC standard, this format has a noticeable advantage over MPEG-1 Layer 3 (MP3), allowing you to achieve better sound quality at the same bit rates. A particularly strong gain is observed in the area of low bitrates: 96 kbps and below. This confirms the promise of the AAC format for digital broadcasting.
The popularity of AAC for music distribution on the Internet today remains low compared to the MP3 format. Users continue to prefer the better portability of MP3 over the stronger AAC compression. A significant part of the music archives on music distribution sites is already natively in MP3 format, and providers do not have access to uncompressed recordings. This means that it doesn't make much sense to recode such recordings into AAC format - the quality is often already lost. However, new pocket players and some online stores already support the AAC format, often with content legality verification (which also discourages users who prefer not to limit themselves to copying music).
Although very promising, the AAC format is not the only highest quality audio compression format. At high bitrates (above 128 kbps), AAC is often inferior to Ogg Vorbis and Musepack encoders. At the lowest bitrates (less than 32 kbps), AAC can be inferior to parametric audio encoders, including specialized encoders for speech compression. However, in the range of medium-low bitrates, AAC currently retains the palm.
Alexey Lukin
Magazine "Sound engineer" 2008 #1
Share
