Simple audio processing to improve older recordings e.g. the Bible

[jensend7]

Some audio content, notably the audio Bible, appears to have been transferred from relatively low-fidelity originals (tape?), has a fair amount of background hiss, and could benefit from some fairly simple processing. (The chapter introductions seem to have been recorded more recently and don't have any near as much background hiss; the noise level difference at the transition between the two can be distracting.) Removing ~12dB of noise using a decent noise reduction filter (I simply used the spectral noise gating filter in Audacity) makes the Bible recording much easier to listen to.

While removing the noise benefits all listeners, it has an additional side benefit for those listening to lossy compressed versions (such as the 32kbps mp3s the Church offers for download): removing noise makes the audio easier for lossy encoders to encode, increasing quality at any given bitrate. (Fewer bits spent encoding noise -> more bits available for what matters.)

Also, since these recordings have very little useful frequency content above ~7.5kHz, they could be encoded with a 16kHz sampling rate rather than the present 22.05kHz. Combined, these two measures give a noticeable increase in the quality of a 32kbps MP3 above and beyond the improvement that noise reduction would accomplish for CD listeners. (Using other codecs esp. Opus would allow for further quality improvements, but I'd better save that for another post.)

The noise reduction and re-encoding would be a fairly simple batch job for anyone with a copy of the masters.

[russellhltn]

There are inexpensive commercial programs. I used one on a musical album and was very pleased with the results. It also uses something like spectral noise gating, but I think it cuts the signal down into narrower bands then what is available in Audacity.

[jensend7]

No need to use commercial programs unless we're sure we've found one that's actually superior (and we aren't already satisfied by results we're getting from free tools). Audacity already allows you to change a frequency smoothing factor to have "bands" as narrow as will work with their transform size (I think they use 2048-wide windowed FFTs, but I'd have to check). Since the tape hiss etc is broadband noise (much like pink noise/Brownian noise), narrower bands aren't that much of a plus. And Audacity's

GWC's filter has a good reputation, but Audacity's might have pretty much caught up to it (perhaps I should test this), and I don't think GWC is operable in batch mode. I would just say to use the effect in SoX since it's ultra-convenient, but SoX's noisered effect seems to have problems with "metallic" artifacting, possibly a problem with attack&decay time/ hysteresis.