What DSPs and in which order for reduce Tape Hiss?
Reply #8 – 2009-10-20 23:22:25
Is not a good idea, I don't want to modify the sound source. I can't see the problem. Don't over-write the original, just create a new file, perhaps in a separate folder, which is a dehissed version, and possibly then prevent the original from being part of the fb2k library/album list (i.e. exclude C:\Music\OriginalHissyTapeTransfers\*.flac or whatever) so you only have the best-sounding copy in your playlists. The de-hissed version doesn't even need to be a large lossless file so keeping the original + dehissed versions needn't double your disk storage requirements. (Any "transparent" lossy format for the dehissed music should be great to listen to, be it something like lossyWAV|FLAC, WVlossy, LAME -V2, AoTuV Ogg Vorbis -q6, neroAACenc -q0.5 or whatever your preferred players can use). If your tape captures didn't utilise Dolby NR but it was utilised on the recording (you can generally be pretty sure they did not if you're hearing bright high-frequencies and lengthened cymbal and hi-hat decays in the hi-frequency part of the spectrum compared to CD or vinyl versions as well as fairly substantial amounts of hi-freq hiss), I'd suggest that Tape Restore Live outputting to a new file with Winamp's Disk Output feature would be an excellent first step. Then you can use fb2k or whatever you prefer to compress the PCM WAV that results to your desired format and then to copy the tags from your original file onto the resulting dehissed file. You may then delete the intermediate PCM WAV and keep just the original tape capture (with hiss) and the dehissed file. If a better dehissing method becomes available, you can always reprocess the original tape capture and create a new dehissed file. Having tried a few audio editors' free/bundled digital noise reduction routines on tapes that I'd captured with Dolby-B NR turned on, I'm of the conclusion that: • The hiss remaining after you have applied Dolby NR is usually masked from audibility by moderate-to-loud music, so digital noise reduction processing is likely to cause more damage than good (e.g. warbling 1960's sci-fi computer sounds/underwater burbling and mucking up the transients in the original). • It's only during fade-in, fade-out, possibly track gaps and quiet tracks or movements that it's seriously worth considering, and then applying to the minimum degree you can get away with to minimize those nasty side-effects that you might not immediately notice, but can become permanently annoying once you recognise them. • There's probably a scope for a psychoacoustically aware digital noise reduction algorithm to be made which automatically determines the masking level and thus would only attempt to de-noise where the noise (measured from the noise-only sample) is predicted to be audible, and would take care to detect transients and avoid smearing them or mistaking them for noise. • I haven't ruled out that other (e.g. professional DAW-grade denoisers or more recent algorithms) are far more transparent, though I suspect that tape noise remaining after Dolby NR on studio master tapes is considerably less perceptible than on consumer cassettes anyway. I did also try an experiment some years ago in arithmetically adding a sample of captured tape-hiss in silence (captured with Dolby B NR turned on) to a sample of ripped CD audio so that I could ABX or ABC/HR the original clean CD version against the dehissed version. The methods I tried weren't very successful at recovering the clean audio from the noise, having robbed it of some of the transients in particular. Given that the noise was moderate, the noise+hiss version (after Dolby B NR, of course) was better than the result of any of the Digital Noise Reductions I tried. For that experiment my recorded tape hiss was perceived to be about 48 dB quieter than the album itself (based on Track Gain and Album Gain measurements, album gain being -0.89 dB). In that experiment the quietest track had an especially quiet intro, with Track Gain of +10.46 dB, meaning 78.54 dB SPL perceived loudness (just measured on the intro, not the whole track, using the Replay Gain formula). The perceived hiss loudness was thus about 37 dB below the perceived intro loudness and was much more objectionable at this time than for the rest of the album (except inter-track gaps and fades), particularly because of relative silences between notes and phrases in the music as much as the perceived overall loudness.