Hacker News Reader

Comment by newaccoutnas | Hacker News Reader

Umm, Nyquist-Shannon Theory? Listening to 24Bit 192Khz can actually damage the cillia in your ear

Listening to 16bit 44.1 kHz can also damage the cilia in your ears. Whether it actually does, in both cases, of course depends on the recorded material and the playback gain.

Bingo! Exactly, Nyquist theory. But no one can make a truly good, brickwall filter at this frequency. It is much easier to make a lot less steep frequency response, whilch would start at 22 khz and would slowly go down all the way to 96 kHz. By the way, how would it damage the ears if it was thoroughly filtered?

Resamplers these days take advance of our faster computers to use near-brickwall filters, like a 1024-point windowed sinc. With this, you can create a steep filter from 20-22 kHz with better than -100 dBr attenuation beyond the Nyquist frequency. This is enough to stop worrying about the aliasing caused by the bandpass filter.

I've never heard of what newaccoutnas is talking about, but I guess if a high-frequency sound is ridiculously loud, the energy could be absorbed somewhere in the cochlea. I've only heard of experiments of subsonic hearing damage, not supersonic.

Well, there are several things to keep in mind. 1. The sharper the frequency response the worse ringing in the phase response. 2. Human hearing is a mystery and the near ultrasonic freq may actually have some influence on the perception of the music. 3. You need to use analog filters before the ADC and after the DAC anyway, and it is so much easier to implement with a much less steep freq response, if used with higher sample rate. I mean, you sample it at 192 and the filter it down to remove anything above 96. There is no point to resample all the time up and down.

That's true for IIR filters but not FIR filters, which modern (This century) signal processing uses. A 1024-point windowed sinc will have linear phase across the entire passband, that's one of the main reasons we use FIR filters.

Do you know what frequency is typically for the resampling? Anything like 192kHz?

Do you mean what sample rates mixing/mastering engineers work in? They're usually 44.1 or 48 times 1, 2, or 4. Software/hardware DSP effects sometimes oversample interally up to 16x. If you're asking what the ADCs use, they usually have an analog filter circuit for each input, sample at 4 or 8x to digital, and then digitally decimate to 1x, all in one $5 ADC chip.

Except that audio DACs are generally oversampling; for example, an output filter for the AD1955 DAC is only needed to reach -3 dB at 100 kHz, even if you are feeding it a 48 kHz PCM signal. This allows these DACs to produce essentially zero energy in the frequency band from Nyquist cutoff of the input signal to filter cutoff.

This resampling adds some latency, right? Ideal resampling adds infinite latency---the sinc function has infinite support.

Yes, but it's not much. To stay with the example given above, group delay is less than 1 ms at 48 kHz fs, largely indepedent of oversampling factor (2-8).

Reply on news.ycombinator.com