Lately I've been transferring my old tapes onto my computer and also recorded several radio theatre programmes using my FM tuner. I have rather good quality hardware, an M-Audio Delta Audiophile 2496, a Nakamichi Cassette Deck 2 and a Technics FM tuner.

The tapes don't seem to have much information beyond 16kHz. The stuff that is visible there when viewing on a frequency analysator in Audition, is probably just hiss. The FM radio captures do have a spike at ~19kHz and somewhat raised levels right next to it on both sides (pilot tone for stereo broadcasts??).

I have captured the audio at 96kHz, 32bits, but as I end up at 44.1kHz before encoding to the final format (which is either AAC or FLAC depending on the case), I was wondering whether I should get better quality if I captured directly to 44.1kHz?

Thanks for any advice.

It does beg the question (from me at least) as to why you aren't just using the LAME MP3 encoder in VBR mode at -V3 to encode the recordings at their original 44.1kHz sample rate.

This would give you the double advantages of files that are only as large as they need to be to capture the available audio content and a cut-off frequency below the 19kHz pilot tone on the FM radio recordings.

Just a suggestion.

Cheers, Slipstreem.

There is no "original samplerate", and I've been wondering whether oversampling and then downsampling is a better idea than going straight to the final sample rate. Probably the difference between the two methods is quite small but I'd still like to take the optimal route while I'm at it.

I've also wondered whether the area near 19kHz has some actual audio content in those FM stereo broadcasts. That is, should I filter it away (or let the codec use the cut-off frequency) or try to keep it in.

You should be able to clearly see from spectrograms that audio ends at 15 kHz.

I've only recorded radio at 48000 Hz, and wonder if more HF information can be captured HF beyond the pilot tone, such as the first stereo difference sideband, or any digital streams sitting on top of mono FM.

Edit: typo

Legally no audio information may be broadcast above 15 kHz. The stereo difference signal is encoded as sideband-only so it wouldn't sound like anything you would want to listen to even if your radio didn't filter it out. I'm not sure about digital, but the old analog secondary audio channel was at 63 kHz, so way above what even 96 kHz sampling could capture.

Edit: typo

If the resampling software is good enough, you will not "get better quality if I captured directly to 44.1kHz" but you will not get worse quality by recording at your target sample rate. You gain nothing by recording at the higher sample rate.

I read this article: http://en.wikipedia.org/wiki/Aliasing. Please explain: does soundcard filters out any frequencies above 22.05 kHz when target sample rate is 44.1 kHz? If not, some high-frequency noise (if any) can become audible.

It all depends on the soundcard. A good one will always apply analog filtering as needed to prevent aliasing. Since many soundcards only input at one frequency, the input filter can be fixed.

Some soundcards only input (and output) at 48 kHz and convert digitally between that and 44.1 kHz.

Soundcards that truly have multiple input frequencies could technically filter anything above 22.05 kHz, since those frequencies are inaudible anyway, but that would seem to defeat the purpose of higher sampling frequencies wouldn't it.

A really good soundcard might always sample at a higher frequency (such as 96 kHz), requiring relatively little analog filtering, and then convert to the target frequency.

You would need to look at the specs of your soundcard, and even then it might not be easy to tell.

If all this is correct, then I should just record the FM broadcasts at 32kHz since there is no audio content beyond 15kHz or so. I would expect my soundcard to be able to record at multiple frequencies properly, but one never knows. I might just ask the M-Audio folks, if it is true, I'm sure they'll be happy to tell me that.

Thanks, pdq.

I'm not sure that I would go that far. If you ever intend to record these files to CD then you should definitely use 44.1 kHz. Even if they will stay forever as mp3 files, lame is primarily tuned for 44.1 kHz and will probably give better results at that sample rate. You might save a little space using 32 kHz, but if you use a 15 kHz or so low-pass filter when encoding, lame will not be wasting bits on frequencies above that.

A good point there. Too bad there is no lowpass filter to tweak in Nero Digital's AAC or FLAC/FLAKE..but I guess I could pretty much do the same trick by filtering everything beyond 16kHz to (almost) zero before encoding.

Aliasing happens at every sample rate -- if there is any input above the Nyquist limit to begin with. With test signals, the results are easy to observe in the proper software. A generated sweep tone going from low audio frequencies to 44+kHz, when recorded at 44.1kHz, will produce an image almost down to zero Hz, although it is very weak after the first couple kHz below the 22kHz cutoff for 44.1kHz sampling. The same thing happens at higher sample rates.

This was consistent on a number of different soundcards I’ve been able to test (including some decent quality cards from M-Audio and Echo). A few other people, stimulated by my posts is several forums, tested other soundcards the same way and reported the same results (more expensive cards than those to which I had access, by chance) . You should be able to find two or more threads discussing this in this forum

However, with real music, especially when the source is LP or cassettes, there is not enough of anything at higher frequencies for you to ever detect aliasing. If you are simply paranoid, regardless of any evidence you can procure on your recording quality, record at 88.2kHz. Resampling to 44.1 is clean, and much faster, than when resampling from a rate that is not an integral multiple of the target rate. This won’t eliminate aliasing in the final product (again, if there was any relevant input from which to produce an image) but the strongest part of the image will be above 42kHz, which is completely eliminated when resampling (good resampling software eliminates the image. Much software isn’t that good.) Especially observe the Goldwave sweep results, but others are also poor.
http://src.infinitewave.ca/

I’ve put many hundreds of hours of spoken audio to mp3, first resampling to 22050kHz. LAME results are excellent. You can also set LAME parameters to lowpass 44.1kHz sample rate files at 11kHz (the cutoff for 22050kHz sample rate) and get only slightly larger files, but I prefer my own processing.

mp3 has more pre-echo at 32kHz than 44.1kHz, since the block sizes are a fixed number of samples, rather than a fixed time period.

You should probably low pass filter at 16kHz whatever you intend to do, and should certainly low pass filter before psychoacoustic encoding - anything preserved above 16kHz from stereo FM radio is junk that wastes bits.

Some very good cassettes can preserve some useful frequency information a little above this (e.g. a couple of kHz higher). Some very good LPs can preserve some useful information quite a little higher (e.g. up to 10kHz).

Irrelevant for me now !

Cheers,
David.

Thanks to everyone for your input. I think that from now on, I'll record at 44.1kHz so that I won't need to use any time for resampling. I just need to apply a lowpass filter and then convert from 32 to 16 bits before encoding to the final format.

I have to disagree with David slightly over his claimed limit on cassette recording upper frequency limitations. The TEAC V-1050 cassette deck that I currently own is perfectly capable of recording to Metal-formula tapes at frequencies of up to 21kHz, as did my previous Technics cassette deck.

Although many manufacturers seemed to struggle to get much past 18kHz, it's not rocket-science to build a cassette deck that works properly and exceeds the human hearing range for minimal cost. It's just a matter of the design engineers understanding how tape recording actually works and putting it into practice properly.

Unfortunately, even many of the big-name manufacturers employed electronic design engineers who seemed to know next to nothing of the theory that underlies good design practice when it came to making cassette decks. I've pulled many lack-lustre decks apart over the years and made modifications costing pennies and gained several kHz at the top end on numerous occasions.

Sorry for dragging the thread off-topic, but properly designed analogue Hi-Fi equipment is one of my life-long passions and I couldn't help jumping to the defence of the much-maligned cassette tape standard.

Cheers, Slipstreem.

While it is true that some cassette decks had much better than average high frequency capacity, unless a tape was made on such a deck, using expensive tape that is capable of capturing those frequencies, and from a source that had those frequencies, and is still in very good condition, that capacity is all irrelevant to transferring cassettes to digital media. Few commercially produced cassettes met those criteria when they were new.

Yep. No argument there.

Cheers, Slipstreem.

I don't suppose you can record a linear tone sweep from, say, 20Hz to 22kHz, play it back, and post the resulting spectrogram?

I don't doubt it does what you say, but I'm interested to see what it does along the way.

Cheers,
David.

I'd be very interested to see the results of such a test too, but I don't have the necessary test equipment to carry out such a test. If you know of any freely available software to output said test tone sweep from a PC line-out and then read it back in through the line-in and create a spectrogram then I'll happily do it and post the results.

A test on this particular Teac would interest me more than the results from any other cassette deck I've ever owned as it's the only one out of three top-flight decks I've had that didn't need any recalibration to get it to meet its specified performance figures.

I appreciate that even though a cassette deck can sound very nearly perceptually transparent, the output when measured using test equipment will undoubtedly stray far from perfection. It would be interesting to see just how awful this one looks on paper.

Cheers, Slipstreem.

Audacity can do exactly what you want with very little effort. Just use "chirp" in the "Generate" menu, then play with one Audacity window and record with another. Then edit->select all then "Plot Spectrum" in the "Analyze" menu. It's not the "best" way of doing this kind of measurement, but is very likely to give you a good idea of the frequency response of the system.

Using noise instead of a chirp will probably work well, too.