Volunteer@Home

Not so far ago Apple announced the initiative aimed to form a kind of standard for mastering of audio content intended for distribution through iTunes Store. Mastering guidelines for audio engineers and independent musicians as well as a set of special audio tools are available on Apple site. The guidelines could be summarized as follows:

• If possible master recordings should be submitted to iTunes Store as 96/24 audio files; remastering is advised in cases when it could provide discernible improvements.
   
• These high resolution files must be free of clipping; it is recommended to have at least 1dB headroom for peak levels.
   
• Apple highly recommends to preserve full dynamic range of recordings which is usually well defined by creative purposes of the author. Over-compression of dynamic range (“loudness war”) is no longer a way to attract listeners because almost all distribution and broadcasting channels now have technologies to control playback volume regardless of actual loudness of a track. Itunes has them for sure.
   
• As almost all tracks in iTunes Store are distributed now in iTunes Plus format (44.1/16, stereo, AAC, VBR-constrained, 256 kbit/s) it is recommended before submitting to preview audio material in the above format and to listen it on various consumer devices using both headphones and speakers.

In order to ease following these guidelines Apple provided a set of audio utilities for Mac OS:

afconvert performs sample rate and bit depth conversions as well as AAC encoding exactly the way iTunes Store does. So any author has possibility to preview quality of final AAC file which will be distributed. Conversion from high resolution files into AAC is made through intermediate 32bit .caf file which helps to avoid dithering thus the whole process of encoding preserves the maximum possible quality.

afclip finds all possible clippings in audio file including the ones between samples; it is recommended to test audio material for clipping both before and after encoding as the latter could introduce additional clipping sometimes.

AURoundTripAAC Audio Unit is intended mostly for blind ABX testing of encoded audio material. With this tool an author can test critical parts of the track for encoding artifacts audibility. No doubt this utility will be helpful also for those listeners who would like to test their ability to hear coding artifacts at various bit rates.

Master for iTunes Droplet automates some of the above procedures.

First AAC@256 encoded audio tracks appeared in iTunes Store in 2007. Soon after that SoundExpert added AAC@256 encoder from the latest at that moment iTunes to the rating system, though there was no confidence that audio material for iTunes Store prepared with that exact encoder. Now Apple disclosed the whole procedure of encoding. The encoder itself was improved since 2007. Taking all this into account we decided to encode SE test files into iTunes Plus format using Apple mastering guidelines and add them to the 256 kbit/s section.

Encoding was performed on Mac OS X 10.7.3 (11D50b) using two recommended commands:

afconvert test.wav intermediate.caf -d 0 -f caff –soundcheck-generate

afconvert intermediate.caf -d aac -f m4af -u pgcm 2 --soundcheck-read -b 256000 -q 127 -s 2 final.m4a

Decoding:

afconvert final.m4a decode.wav -d LEI24 -f WAVE

Resulting 24bit .wav file was used to prepare SE test files as usually, thus preserving the highest possible audio quality the iTunes Plus format can offer for 44.1/16 source material.

Out of curiosity SE test files were also encoded with the latest iTunes for Windows (10.6.1.7) using iTunes Plus setting. Difference levels for both encodings are in the Table 1. 

Table 1: Difference levels [dB] of SE test files encoded into iTunes Plus format on Mac OS and Windows.

Similar values with small variance between encoders mean that they use the same psycho-acoustic model but differ in computation procedures. Usually such small variance of Df levels is imperceptible and it can be safely assumed that both encoders produce audibly identical files. So anyone with the latest iTunes installed can encode their music with the exact quality established in iTunes Store.

Mastered For People

Apple's initiative to create some framework standard for digitally distributed music seems actual and praiseworthy. Exactly poor mastered recordings is the main source of “bad sound” problem nowadays. Using of psycho-acoustic encoding on the final stage of distribution just worsens the problem not originates it. Even taking into consideration that iTunes is only a small part of global music distribution, Apple is capable to influence on mastering practices worldwide as the company is well known trend-setter today. Not to mention the sweet marketing goal for the company to give the name – Mastered For iTunes – to this new trend of more careful mastering practices.

On the other hand iTunes Store is only a tiny, almost insignificant part of global music sharing on the internet. P2Pnetworks, cyberlocking and cloud services, social networks and author's personal web sites rapidly become effective channels of music distribution. Distinction between distribution and sharing become less and less obvious. Definitely some mastering guidelines are necessary for authors and musicians who get on in the world this new way. Apple's guidelines with a few remarks could serve the purpose and such “Mastered For People” recommendations may look like follows:

1. If possible retain your audio tracks as multi-track projects for

• further improvement and development
• producing various master records in 2.0, 5.1, 7.1, 3D … formats
• sharing your work as multi-track mixes for “creative listening” - making by listeners their own versions of your tracks

2. Retain at least 96/24 stereo mix with full dynamic range and 1dB headroom for peak levels (-1 dBFS)

3. While making equalization, limiting, dynamic range compression ... of the 96/24 master for final distribution:

• use 32bit mode
• avoid clipping (-1 dBFS)
• lo-fi versions of your track may require special post-mastering
• resulting master record should be 96/24 as well

4. Variants of conversions for final distribution/sharing:

[96/24 master] → downsampling → [48/32] → dithering → [48/16] → FLAC | ALAC | RAR | …
[96/24 master] → downsampling → [48/32] → Vorbis | AAC | MP3 | … @256-320 kbit/s
[96/24 master] → downsampling → [44.1/32] → dithering → [44.1/16] → Audio CD

Sample frequency 44.1kHz seems become more and more obsolete/senseless as it helps only to solve some compatibility issues such as creation of Red Book Audio CDs and using of old audio equipment. Sampling frequency 48kHz serves well both professional and consumer audio today making 44.1kHz unnecessary.

The draft is open for discussion, drop us a line if you are interested.

SoundExpert thanks James Orlov for encoding test files on his perfectly updated MacBook Pro and for technical assistance with Mac OS X.

As usually SoundExpert encourages you to take part in listening tests. All SE ratings, including iTunes Plus 2012 ones, exist only thanks to visitors like you. Testing is short and easy, visit our Testing Room for simple 1-2-3 instruction.

Please, remember, downloading a test file from SoundExpert you will not necessarily get a test file of iTunes Plus 2012. SoundExpert testing is blind testing – you don't know the device you test. This is the only reliable way of getting true audio quality ratings, free from any cheating and various human senses other than perception of sound quality. After you have sent your grade you will see the device you tested.

Best regards,
SE Team.

According to Xiph.Org new release of popular Vorbis audio codec was out on 2012.02.03. It contains only bug, doc and security fixes (no quality improvements). Previous version – 1.3.2 – was released on 2010.11.02.

The latest Vorbis encoder was added to 256 kbit/s section with the following settings:

Vorbis VBR@256.4 (Xiph 1.3.3) - Ogg Vorbis, 256.4 kbit/s FBR
CODER: OggEnc v2.87 (libVorbis 1.3.3)
- usage: oggenc2 -q8.2 ref.wav
- 44100 Hz Stereo
DECODER: OggDec v1.9.7 (libVorbis 1.3.1)
- oggdec ref.ogg

As usually encoding was made with OggEnc utility and decoding – with OggDec utility without dithering.

What is Vorbis codec

Ogg Vorbis is a fully open, non-proprietary, patent-and-royalty-free, general-purpose compressed audio format for mid to high quality audio and music at fixed and variable bitrates. It means that you can freely encode and distribute your music in high quality without paying to anyone.

The Vorbis bitstream format was frozen in 2000 and all bitstreams encoded since will remain compatible with all future releases of Vorbis.

Extensive FAQ about Vorbis format is on Xiph.Org.

Where to download Vorbis codec

Vorbis encoder, decoder and other tools are available for all major platforms. Visit Vorbis.com to get them.

SoundExpert thanks David for this codec suggestion. We also received other suggestions on the codec and plan to add it to different SE bitrate sections. Your proposals are still welcome.

Extensive research experiments with Galaxy S revealed that accuracy of test files preparaton for analog audio devices must be higher than we used for iPhone 4. As a result comparison of players could be unfair. We decided to prepare test files for iPhone 4 from scratch using our latest recordings of white model with iOS 4.3.5. Unfortunately there is no possibility to account the grades submitted with old test files. They were discarded, sorry!

New test files are alredy in the system. Article about iPhone 4 was corrected accordingly.

Best regards,
SE Team.

Figure 1: Galaxy S smartphone by Samsung

This smartphone by Samsung is considered by many as a competitor of iPhone 4. So we desided to add it to our new "Portable players" section side by side with its rival. The model we tested is GT-I9000 with Android "Gingerbread" operating system. Our main conclusions in short:

• Objective measurements show that audio tract of Galaxy S is on a par with similar devices of this class. Final conclusion about  its quality can be made only after finishing listening tests when its rating become reliable. Consider taking part in SoundExpert tests.
  
  
• If you listen it with wired headphones and satisfied with the loudness, you don't need to upgrade original audio drivers to Voodoo ones. If you upgrade though, you'll get additional output power without quality sacrifice.
  
  
• Besides common for all audio players noise and gradual cutting of high frequences Galaxy S has some special flavour of sound – spatial distortion of stereo image. But the level of the degradation is too subtle to be noticed even in perfect cans.

Objective measurements

First of all we failed to get our standard required output level from headphones OUT of Galaxy S, connected to Koss Porta Pro. Even at max volume available with original sound drivers we fell short of 2dB SPL. Actually it's not a big problem for the objective measurements but limitation of output level is obvious and can be a problem for high impedance or low sensitivity headphones.

Difference levels for our standard set of signals and samples are in the Table 1:

Table 1: Difference levels of various signals and samples transfered through Galaxy S with naitive sound drivers. The values show to what extent the waveformes change their shapes passing through device:

-∞ dB - signals passed unchanged, perfectly exact shapes
0 dB – input and output signals have nothing in common

For sinusoidal signals and small Df values Difference level can be easily converted to THD+Noise:

THD+Noise [dB] ≈ Df [dB] + 3 [dB].

Sound pressure levels of samples and signals are unweighted and averaged over 400ms.

The values look good and depend on complexity of signals; nothing extraodinary. If you like to listen the sine wave @1kHz at the output of Galaxy S with artifacts amplification +71dB refer to Audio 1.

Preview this Audio in lossy format (vorbis@160 or mp3@165 depending on your browser) Download this Audio in lossless format (10sec, FLAC:1.24Mb)

Audio 1: Sinusoidal signal (@1kHz, 86dB SPL unweighted) at the output of Galaxy S with sound artifacts amplified to +71dB

The sine signal is deeply buried in noise. The origin of its audible fluctuations is unclear. Our guess – it is headphones amplifier of class W at work. Galaxy S uses Walfson WM8994 audio hub with integrated power amplifier of this class. Such amplifires dinamicly adjust voltage level for their output stages maximizing efficiency and reducing heat dissipation this way. The adjustments are made discretely depending on power required at each moment. So what we hear is probably some modulation of signal by those adjustments fluctuating around optimum voltage value (as the signal level is costant). If somebody can suggest better explanation, please, drop us a line. And yes, this pecularity doesn't depend on audio drivers, software players, input files (WAV or FLAC) and specific to Galaxy S only.

Also, despite heavy isolation of headphones from background noise during recording, some of our voices are noticable at this level of artifacts amplification. We'll find who cried louder than others during that particular recording and punish demonstratively. All future recordings will be made in silence. Promise.

In any case this oddity at the level of -70dB doesn't affect real-life listening experience but can be interesting for developers of amplifires.

Voodoo audio drivers

Opensource project Voodoo originated by François Simond aka supercurio from xda-developers forums developed alternative and very popular audio drivers. They unlock full power of headphones amplifier and make possible some low level tunings of WM8994 chip which improve sound quality. For assessment of those improvements we installed Voodoo Control Plus 3.0 (WM8994 v10) with the following settings (Figure 2):

Figure 2: Settings of Voodoo sound drivers we used

Now we could easily achieve required sound pressure level. Results of our standard measurements are in the Table 2.

Table 2: Difference levels of various signals and samples transfered through Galaxy S with Voodoo sound drivers. The values show to what extent the waveformes change their shapes passing through device:

-∞ dB - signals passed unchanged, perfectly exact shapes
0 dB – input and output signals have nothing in common

For sinusoidal signals and small Df values Difference level can be easily converted to THD+Noise:

THD+Noise [dB] ≈ Df [dB] + 3 [dB].

Sound pressure levels of samples and signals are unweighted and averaged over 400ms.

The improvements are not dramatic. Audible difference is also very subtle and illustrated by the piano excerpt (Audio 2).

Preview this Audio in lossy format (vorbis@160 or mp3@165 depending on your browser) Download this Audio in lossless format (15sec, FLAC:1.21Mb)

Audio 2: Piano excerpt played back on Galaxy S using Voodoo (1) and original (2) sound drivers. In both samples sound artifacts are amplified to +32dB. The second one (with original sound drivers) has a bit more noise and more "dirty" high frequences because they are reproduced less accurately. Though the difference is very subtle even with this huge artifacts amplification.

Taking into account that quantative improvements, measured with Diff. Level, are within 0.1dB for real-life signals and audible difference is faintly discernible, we can safely conclude that such improvements are far beyond the threshold of human perception. Original sound drivers in this sense are similar to Voodoo ones. So if there is no need for additional output power from your Galaxy S (very reasonable self-restriction nowadays), there is no necessity to install Voodoo drivers. In case of using some headphones with low sensitivity or high impedance or both Voodoo drivers will help to drive those cans without loosing audio quality.

Sound flavoring

Another feature of Galaxy S sound is some subtle flavoring caused by small and unusual cross-talk between stereo channels. It is unusual because amplitude and phase of signal penetrating into neighboring channel are frequency dependent. In most cases penetrating signal consists of middle-to-high frequencies, has inverted polarity and amplitude -40dB (and lower) below level of main signal in channel. In order to assess consequences of such cross-talk we prepared two stereo recordings – one usual with cross-talk and one without. For the purpose the piano sample, played back by Galaxy S, was recorded three times – left and right channels separately and in stereo as usual. Then left and right channels were combined in stereo thus completely eliminating any cross-talk in this stereo recording. In both stereo recordings sound artifacts were amplified to +16dB. Audio 3 presents result of the experiment. It consists of three samples – original piano excerpt, played back version without cross-talk and version with cross-talk.

Preview this Audio in lossy format (vorbis@160 or mp3@165 depending on your browser) Download this Audio in lossless format (23sec, FLAC:1.42Mb)

Audio 3: Three sound samples show the influence of cross-talk in Galaxy S on resulting sound quality. First sample is a reference piano sample. Second – the same sample played back by Galaxy S without cross-talk (left and right channels were recorded separately). Third sample was played back and recorded in stereo normally thus it is affected by cross-talk. Sound artifacts of two last samples were amplified to +16dB making the difference between samples more audible.

In the third sample (affected by cross-talk) sound image is spread among channels as if it would been processed by spatializer. If it were not for so low levels of this cross-talk we would suppose it was made in purpose. Cross-mixing of stereo channels with inverted polarity is often used for widening of stereo base in various sound enhancers. But in this case we rather suppose it is a side effect of analog circuitry design, again the headphones amplifier most likely. The effect becomes audible only with pretty high levels of artifacts amplification (+12dB and more). At these levels increased noise and absence of high frequencies are already revealed vividly in sound, so the results of this small cross-talk can hardly be heard in real-life.

All changes a sound undergoes passing through Galaxy S can be summarized as follows:

Table 3: Various sound artifacts revealed at different levels of artifacts amplification 

All these types of sound degradation in various combinations (except the last one which is reliably masked by the three first) are present in test files of Galaxy S uploaded to SE testing engine. Listening tests will show how far these sound drawbacks lay beyond threshold of human perception.

As usually we would like to thank some people. First of all Kirill who placed his smartphone to our disposal and patiently performed all reconfigurations and modifications necessary for more than dozen recordings we made using different settings. His Galaxy S was not hurt during those inhuman tests. When most of the work was finished we received suggestion from Alex to include Galaxy S and S II to SE testing. It was encouraging, thanks.

And some bad news at the end. Bad news for iPhone 4. No, it is not a looser in comparison with Galaxy S. Our extensive research testing of the latter revealed that test files of iPhone 4 that already in the system were prepared with not sufficient accuracy. For fair comparison of portable players they must be replaced. This will be done within a few days followed by corresponding notification in SE news. Sorry for that.

SE Team.

Figure 1: iPhone 4 smartphone by Apple

It finally happened, new section "Portable Players" has its first contender – iPhone 4 (white GSM model with iOS 4.3.5 and 8Gb of storage). This “player” was chosen due to a few reasons:

• it is popular
• it has the same audio circuitry as iPod Touch 4 and therefore produces very similar level of audio quality
• convergence of portable players and means of communication seems to be inevitable

This is a beginning, all other players will be chosen according to your suggestions as usual. Procedure of recording SE test signal for portable players is pretty solid now, so we are not excluding possibility to ask help from people outside SE to record SE test signal for some players which we don't have at our disposal, especially as we already had successful experience of this kind with MD-recorders (ATRAC) a year ago.

Here is some basic points of testing conditions for portable players at SE:

• nine SE sound samples are uploaded to player as uncompressed WAV file
• all audio enhancements are off
• headphone's out of player is loaded with headphones Koss Porta Pro
• playback level is adjusted to have some predefined and equal for all players sound pressure level
• output signal (with headphones load) is recorded by hi-res recorder (MicroTrack II at the moment)
• after precise time warping of recorded signal it is treated like any other output signal at SE – sound artifacts amplification, preparing of test files, usual SE listening tests …

(full specs of testing conditions will be available in “Portable players” section)

With these testing conditions we tried to achieve two goals:

• to make sound quality testing as close as possible to real-life usage of portable players
• to make resulting ratings directly comparable to existing SE ratings (of audio codecs)

Testing details specific to iPhone 4

Along with usual nine SE samples we also used sinusoidal signal (1kHz, -12dB RMS) and white noise (Gaussian, -12dB RMS) for some technical measurements. In particular the most interesting are the levels of degradation of initial shapes of those signals. Being transferred through player the signals are distorted to some level which can be conveniently measured with Difference level parameter. Lower Df values correspond to more accurate (less distorted) transfer of signals. The utmost limits of Df parameter are:

-∞ [db] – signal shapes are perfectly identical
0 [dB] – signals are completely different (nothing in common at all)

Table 1. Difference levels [db] of various signals transferred through iPhone 4.

According to the measurements sinusoidal signal is transferred rather accurately (Df ~ -74 dB). For sinusoidal signals and small Df values Difference level can be easily converted to THD+Noise:

THD+Noise [dB] ≈ Df [dB] + 3 [dB].

Resulting value of THD+Noise -69 dB (0.04%) looks good for a portable player loaded with headphones. But if we replace sine with white noise Df drops to -5 dB. In this case we can't convert Df to THD as the latter is defined for pure tones only. Keeping that in mind we can still formally convert Df = -5.6 dB to THD+Noise = -0.9 dB (90%) and say that white noise being transferred through the player changes its initial shape dramatically. All other real-world sound signals are transferred with Df values between -74 dB and -5 dB (SE nine samples are in the table). Such accuracy spread of reproducing pure tones and more complicated signals is not surprising. This is more or less common for all audio equipment and iPhone 4 actually looks pretty good in comparison with other players that we tested.

Specific feature of iPhone 4 sound is less accurate reproduction of highs in comparison with other frequency regions. This is mostly caused by phase distortions. SE testing procedure doesn't allow to discern origin of the distortions - hardware circuitry or software processing but it allows to hear them (Audio 1).

Preview this Audio in lossy format (vorbis@160 or mp3@165 depending on your browser) Download this Audio in lossless format (16sec, FLAC:1.09Mb)

Audio 1: Phase distortions of highs in iPhone 4 amplified. The first sample is referense piano excerpt, the second - output of iPhone 4 with artifacts amplified to +32dB. Outstanding highs reveal phase inaccuracy in that frequency region.

Such inaccuracy could lead to less transparent reproduction of highs, in particular - to bluring of their spatial positions. Resulting effect will depend on how far such inaccuracy is beyond threshold of human perception. No measurements (including artifacts gain) can reveal that, results of SE listening tests are more reliable in this sense as they take into account how any particular bunch of distortions is assessed by human hearing system.

Best regards,
SE Team.