Massimo Procopio

High Definition Audio vs CD


High Definition Audio

Is High Definition Audio (also called Hi-Res, High Resolution) really better than CD quality?

Do 24bit and 192Khz really bring benefits in sampling the music track?

For a long time I have heard and read various opinions about HD sound quality versus the more prevalent CD quality.

I never had the chance to actually test the differences between the two different audio sampling qualities with my own ear until I bought high quality headphones and hardware.

What were results of my tests?

After a first apparent belief that the 24bit / 192Khz format gave more body to the sound, I wanted to do a double blind test, getting help from my wife. The result left me stunned: after a few attempts, not only I didn't understand which was the high resolution file, but, mistakenly, I inverted them, attributing HD quality to the CD format.

So I wanted to use a more scientific approach to better understand what the mind was boycotting by deceiving me with my feelings.

Here are my evaluations, but I would like to point out some of my beliefs first.

The digital signal, therefore sampled, can never be as the analogue one, for obvious reasons (and this is not always necessarily negative), and live music can never be the same as that reproduced by electronic instruments, no matter how much it may be closer to reality. As a musician, I can say with certainty that when you are immersed in the "real" sound of musical instruments, listening is profound, rich in harmonics and vibrations that electronic instruments can hardly ever reproduce entirely. This does not mean that a good audio system can come very close to the pleasure of listening to live music.

That said, let's continue with the analysis focusing only on the differences in quality of the various digital formats.


High Definition Audio vs CD

For the tests I used three songs: Starlight and Supermassive by Muse (24bit / 96Khz), and a 24bit and 192kHz classical music test file.

To get the most scientifically valid result I uploaded the files to Audacity, and I manipulated them as follows:

  1. Conversion from 24-bit 96Khz (or 192Khz) to 16-bit 44Khz (or 48Khz);
  2. Making of a new file with two stereo tracks in Audacity. In the first I inserted the high definition file, in the second the converted one;
  3. Lower resolution waveform inversion;
  4. Sum of the tracks and save in a suitable format.

In doing so, I tried to subtract the two tracks to obtain a final file with only the differences in the audio content.

These are the results with the relative frequency spectra:

High Definition Audio Analysis
Starlight - Difference of audio content between 24b / 96Khz and 16b / 44Khz
High Definition Audio Analysis
Supermassive - Difference of audio content between 24b / 96Khz and 16b / 44Khz
High Definition Audio Analysis
Supermassive - Difference of audio content between 24b / 96Khz and 16b / 48Khz
High Definition Audio Analysis
Classical Music Track - Difference of audio content between 24b / 192Khz and 16b / 44Khz

As you can guess from the graphs, the difference is practically zero. No information remains.

By amplifying the track obtained from the differences of the two Starlight formats by 50 dB we obtain this result:

High Definition Audio Analysis
Starlight - Difference of audio content between 24b / 96Khz and 16b / 44Khz amplified by 50 dB

Below are the audio files obtained as a result, compressed in mp3 to 320k (The wav format would not be feasible on an internet page due to its excessive size, but I can assure you that the result does not change, since the track contains absolutely nothing except noise):

Starlight-24b-96k-Diff-16b-44k

Supermassive-24b-96k-Diff-16b-44k

Test_Classic-24b-196-Diff-16b-44k

Starlight-24b-96k-Diff-16b-44k-Amp50db

As you can hear, there is no useful audio information. At this point it is possible to say that the high resolution format does not bring any real benefit. Rather it is necessary to evaluate the various compression formats to understand what the real audio information is being lost.

Furthermore, in the various tests carried out I used various conversion software for Wav from 24 to 16 bit, from 192Khz to 44Khz, and not all of them guaranteed a result as precise as the one obtained with Audacity. Conversion errors, however, could be considered practically nil.

Here is an example:

High Definition Audio Analysis
Supermassive - 24b / 96Khz - 16b / 44Khz audio content difference obtained with SwitchConverter

It is difficult to see the differences from the image, but in reality the track is not completely flat like that obtained with the conversion done with Audacity, even if the information lost is negligible. Of all the converters used, excluding Audacity, Switch Converter proved to be the best anyway.

The result obtained with the conversion to other formats is different.


High Definition Audio vs FLAC

I made a comparison between two formats: FLAC and MP3.

Flac is a lossless compressed format, on the contrary mp3 is a lossy compressed format.

In the first case the results obtained are the same as those obtained with the wav. In fact, there was no loss, and the results were identical, confirming the audio quality of the format intended for audiophiles.


High Definition Audio vs MP3

As for the mp3, the result was completely different, and even unexpected. The loss of audio information is far from negligible.

Here are the results:

High Definition Audio Analysis
Supermassive - Difference between audio content from 24b / 96Khz to mp3 320K
High Definition Audio Analysis
Supermassive - Difference between audio content from 24b / 96Khz to mp3 128K

Here are the audio files obtained:

Supermassive-24b-96k-Diff-mp3-320k

Supermassive-24b-96k-Diff-mp3-128k

In this case I preferred to insert only 20 seconds of audio per file to avoid copyright problems, since the information lost is considerable, and from the files obtained it is possible to listen to the entire song (even if with poor quality).

Now a bit of theory to support the results obtained.

Audio resolution depends on two factors: the bits used to sample the signal and the sampling rate.

Let's start with the bits.

24 bits correspond to 2 elevated to twenty-fourth combinations, that is 16777216 possible variations in the amplitude of the signal. 16 bits, on the other hand, correspond to 65536 amplitude values.

The difference between 24 and 16 bits is considerable, but we must consider that a variation in the amplitude of the output signal (i.e. the volume, in simple terms) of 1/65536 is practically impossible to perceive because it is an extremely small variation. In fact, from the previous results, it is possible to see how the difference between 24-bit and 16-bit audio information is practically zero, and even amplifying it by 50dB is completely negligible.

So 16 bits, both from a mathematical point of view and from the results of my tests, are sufficient to guarantee faithful audio reproduction.

The sampling frequency, according to the Nyquist-Shannon theorem, must be at least twice the audible frequency. So if we want to sample from 0Hz to 22Khz 44Khz is sufficient, and this was the mathematical reasoning that led to the birth of the CD format (the audible frequency ranges from 20Hz to 20Khz). The results of the previous tests confirm what was calculated mathematically: the difference between the track at 96Khz (or 192Khz) and that at 44Khz is zero in the audible frequency range.

At this point we can safely conclude that the 24 bit / 192 Khz format does not contain any more useful information than the CD format, and that the compressed Flac format is great for storing audio files at the highest quality without taking up too much space.

The same cannot be said about the mp3 format, which, especially at 128k, compromises the fidelity of the recorded audio signal. With a quality audio system, a particularly gifted ear may actually not fully enjoy the mp3 quality and, indeed, have discomfort in listening.

As I specified at the beginning of this article, what was said only concerns digital formats, because for analogue, such as vinyl, the listening pleasure and the fidelity of reproduction can be completely different, and, probably, differences are not easily quantifiable.

Happy listening to all.

Massimo Procopio

2 Comments

  1. Pietro

    Caro Massimo,
    pur non avendo avuto la fortuna di conoscerti direttamente, mi permetto di darti del tu.
    In questi giorni, complice la quarantena forzata, ho potuto impiegare del tempo nel tentare capire bene la faccenda dell’audio ad alta risoluzione. Mi ero già imbattuto in questo tuo articolo trovando ben fatto e soprattutto scevro da “psicoacustica”.
    Ho ripetuto le tue prove anche con mio materiale audio e tutto coincide. Del resto essendo io un elettronico e conoscendo bene il teorema del campionamento, non avevo dubbi.
    Inoltre ho messo su un sistema di confronto pratico tra sistemi di streaming.
    Hardware: Raspberry Pi+ Dac BB PCM5122 uPNP (S.o. Volumio)
    Software1: Su Macbook Audirvana con Qobuz che invia lo stream via uPNP all’hardaware
    Software2: Su Macbook Spotify Premium che invia lo stream all’hardware, che riceve con plugin Spotify apposito.
    Su Audirvana ho cercato album (su Qobuz) 24/192 che trovavo anche su Spotify (dettato alla massima qualità che credo sia 320 kbps).
    Con questo setup è possibile mettere in pausa su Audirvana e proseguire su Spotify sullo stesso brano quasi istantaneamente.
    Risultato: nessuna apprezzabile differenza, forse qualche dettaglio nei fiati e nelle voci, eccetto il volume diverso che, con gli appositi controlli del volume delle 2 applicazioni, ho cercato di rendere uguale.
    Spero che quanto scritto possa essere da te apprezzato.
    Saluti.
    Pietro

    1. massimoprocopio

      Ciao Pietro. Mi fa molto piacere leggere il tuo messaggio. Grazie a te, adesso ho una conferma in più della validità di quanto ho scritto in questa mia pagina.
      Online si trovano informazioni molto differenti, spinti dalla “psicoacustica” (hai trovato un nome perfettamente adeguato!!) o dagli sponsor.
      Sono convinto che possiamo riempirci di parole, ma solo i fatti parlano veramente. Per questo ho voluto fare un’analisi libera da preconcetti e ricca di dati solidi, e non
      di ipotesi o giudizi personali. Il test che hai fatto tu da ulteriore supporto ai miei, confermando che l’High Definition Audio non apporta migliorie di nessun tipo
      all’ascolto “casalingo”. Ti dirò di più, grazie ai test che ho fatto, sono portato a pensare che quella leggera differenza di dettaglio che hai percepito nei fiati e nelle voci, sia dovuto, più che altro, alla differenza di codifica, come probabilmente anche il volume, che tu, giustamente, hai portato allo stesso livello per poter condurre un’analisi adeguata.
      Grazie per il tuo supporto.
      Se hai dati che vuoi pubblicare e vuoi essere aggiunto come supporter di questa pagina, mandameli e farò una sezione in più con i tuoi test.
      See you soon.
      Massimo

Leave a Reply

Your email address will not be published.

*

en_GBEnglish