As computer audiophiles we love to obsess about 'bit perfectness' ('bit perfection'?), in other words whether or not the integrity of the audio samples being properly preserved in the digital domain. This is particularly relevant when ripping CDs, since a data error that makes its way into the ripped file will remain there forever. However, given that on an audio CD each 16-bit word (some simplification here) encodes one sample representing only 1/44100th of a second of audio it is legitimate to ask whether these errors are generally audible at all. Elk asked this very question in another thread so I decided to take a closer look.
I took track 17 from Stereophile Editor's Choice: Sampler & Test CD (a 1kHz tone at -20dB) and imported it into Audactiy. Then using the 'Draw' tool I edited the waveform at one second intervals to introduce -6dB 'spikes' into both L and R channels. I changed one sample one second into the track, two consecutive samples two seconds in etc. My plan was to see how far into the track I could play before these data errors became audible. Too easy! Even at one second the resulting click was very clearly audible.
Next I experimented with reducing the amplitude of the click to see at what point it would become inaudible. I successively lowered the level of the rogue sample by 6dB (roughly equivalent to one bit's worth of amplitude) at a time to -12dB, -18db, -24dB, -30dB, however it was still clearly if faintly audible even at -36dB. It is worth pointing out that this is a near-worst-case scenario and it is quite possible that a real musical signal would render errors at this level inaudible, however that would necessarily be programme-dependent.
So my conclusion from this fairly simple test is that yes, bit errors can well be audible and that even a single corrupted audio sample can be detected quite easily even if the amplitude of the error is significantly below that of the signal. Given the very low level of errors that can be detected, it is furthermore quite feasible that an error in one single bit would be quite audible as a click on replay, depending on its position in the corrupted word.
Very interested to hear people's thoughts on this, particularly if you think the methodology or conclusion is flawed in any way.
As computer audiophiles we love to obsess about 'bit perfectness' ('bit perfection'?), in other words whether or not the integrity of the audio samples being properly preserved in the digital domain. This is particularly relevant when ripping CDs, since a data error that makes its way into the ripped file will remain there forever. However, given that on an audio CD each 16-bit word (some simplification here) encodes one sample representing only 1/44100th of a second of audio it is legitimate to ask whether these errors are generally audible at all. Elk asked this very question in another thread so I decided to take a closer look.
I took track 17 from Stereophile Editor's Choice: Sampler & Test CD (a 1kHz tone at -20dB) and imported it into Audactiy. Then using the 'Draw' tool I edited the waveform at one second intervals to introduce -6dB 'spikes' into both L and R channels. I changed one sample one second into the track, two consecutive samples two seconds in etc. My plan was to see how far into the track I could play before these data errors became audible. Too easy! Even at one second the resulting click was very clearly audible.
Next I experimented with reducing the amplitude of the click to see at what point it would become inaudible. I successively lowered the level of the rogue sample by 6dB (roughly equivalent to one bit's worth of amplitude) at a time to -12dB, -18db, -24dB, -30dB, however it was still clearly if faintly audible even at -36dB. It is worth pointing out that this is a near-worst-case scenario and it is quite possible that a real musical signal would render errors at this level inaudible, however that would necessarily be programme-dependent.
So my conclusion from this fairly simple test is that yes, bit errors can well be audible and that even a single corrupted audio sample can be detected quite easily even if the amplitude of the error is significantly below that of the signal. Given the very low level of errors that can be detected, it is furthermore quite feasible that an error in one single bit would be quite audible as a click on replay, depending on its position in the corrupted word.
Very interested to hear people's thoughts on this, particularly if you think the methodology or conclusion is flawed in any way.