pre-echo will appear on very fast attacks; clapping, some types of drums and castanets are usually cited as the worst. Some styles of electronic music have fast attacks as well.
With older encoders, cymbals were terrible, but lame's psychoacoustic model is pretty good at masking those artifacts these days (at least at high bitrates).
For examples, some quick searching on hydrogen audio found a couple songs reported to be ABX distinguishable with lame and after previewing the songs, they do in fact have a lot of quick attacks:
Human Disease by Slayer (Some very fast and cleanly played drum parts; also it's mostly not snares; snares sound terrible at low bitrates, but I personally can't distinguish high-bitrate snares from uncompressed snares)
Show Me your Spine by PTP (The "instrument" used for the base rhythm has an unnaturally short attack).
Any sound that starts or end very suddenly. Drums of any kind will sound kind of weird in MP3. Applause used to be the bane of MP3. Encoders got better at it, but it's apparently impossible to get rid of completely. As far as I understand it it's the accoustic equivalent of ringing artifacts that you can see in low quality JPEG's or old MPEG videos.
Playing a set of high-frequency pure sine waves is the failure point for MP3, AAC, Vorbis, and Opus. Dial-up noises are close to this, which you can try encoding/decoding. And this is no surprise, since the point of the V.90/92 protocols is to cram as much information as possible into analog frequencies, and the point of psychoacoustic lossy codecs is to remove the least efficient frequency information of our log-frequency scaled ears.
But this is kind of a pedagogical example. Not the point of who you're responding to.
So, a high frequency DTMF[0] construction should render a pre-echo? What’s the lowest definition of “high” pairs that would satisfy showing this encoding breakage but be hearable by a majority of people[1]?
V.90 and V.92 both sound largely like spectrally shaped noise. Even if you give very regular inputs, the data is compressed and scrambled, then spectrally shaped and then output as 8-bit 8kHz log-weighted PCM.
pre-echo will appear on very fast attacks; clapping, some types of drums and castanets are usually cited as the worst. Some styles of electronic music have fast attacks as well.
With older encoders, cymbals were terrible, but lame's psychoacoustic model is pretty good at masking those artifacts these days (at least at high bitrates).
For examples, some quick searching on hydrogen audio found a couple songs reported to be ABX distinguishable with lame and after previewing the songs, they do in fact have a lot of quick attacks:
Human Disease by Slayer (Some very fast and cleanly played drum parts; also it's mostly not snares; snares sound terrible at low bitrates, but I personally can't distinguish high-bitrate snares from uncompressed snares)
Show Me your Spine by PTP (The "instrument" used for the base rhythm has an unnaturally short attack).
Any sound that starts or end very suddenly. Drums of any kind will sound kind of weird in MP3. Applause used to be the bane of MP3. Encoders got better at it, but it's apparently impossible to get rid of completely. As far as I understand it it's the accoustic equivalent of ringing artifacts that you can see in low quality JPEG's or old MPEG videos.
Playing a set of high-frequency pure sine waves is the failure point for MP3, AAC, Vorbis, and Opus. Dial-up noises are close to this, which you can try encoding/decoding. And this is no surprise, since the point of the V.90/92 protocols is to cram as much information as possible into analog frequencies, and the point of psychoacoustic lossy codecs is to remove the least efficient frequency information of our log-frequency scaled ears.
But this is kind of a pedagogical example. Not the point of who you're responding to.
So, a high frequency DTMF[0] construction should render a pre-echo? What’s the lowest definition of “high” pairs that would satisfy showing this encoding breakage but be hearable by a majority of people[1]?
[0] https://en.wikipedia.org/wiki/Dual-tone_multi-frequency_sign...
[1] https://en.m.wikipedia.org/wiki/Presbycusis
No, pre-echos are unrelated to high-frequency sine waves. And DTMF aren't what I'd consider high-frequency. I'm talking about 4-20kHz.
1 reply →
V.90 and V.92 both sound largely like spectrally shaped noise. Even if you give very regular inputs, the data is compressed and scrambled, then spectrally shaped and then output as 8-bit 8kHz log-weighted PCM.
https://www.youtube.com/watch?v=MtiRBFWkRKs beginning at 54 seconds.
I don't really hear the pre-echo.
Actually I love that I don't hear compression artifacts in music or see them in JPEGs. Makes my gear so much cheaper. :-)