Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: FM Broadcast: 44 kHz or 48 kHz (Read 11439 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

FM Broadcast: 44 kHz or 48 kHz

Reply #25 –
Quote
Isn't a -V preset optimised to give the best possible performance in terms of quality vs bitrate for the bitrate used, ergo, acceptable results if you match the bandwidth of the preset to the source?


Yes, in terms of encoding CDs it systematically discards HF signals in an attempt to gracefully degrade quality on a number of fronts to get the best overall "bang-for-your-bitrate" compromise on average. If the CD happens to contain vintage recordings with no content over 12 kHz and in mono, the bitrate use will plummet, while retaining the same level of transparency that is normally available. So -V2 used on mono 1960s records will often provide transparency at less than 128 kbps.

There's no useful signal from FM above about 15 kHz, and a potentially intrusive 19 kHz pilot tone which will either be audible or bloat your bitrate, possibly both, if present, so it ought to be removed.

Here we know something about the signal that LAME can't know - that there's no intended signal above about 15 kHz, so the lowpass setting (unless it is higher than that of the proposed quality preset!) is providing useful true information to LAME without forcing it to encode frequencies it would be better to discard to help clean up our signal.

The -V settings control far more than lowpass, particularly the allowable signal-to-mask ratios (which have much more effect at reducing bitrate). As you lower the quality target (increasing -V number) you allow more error between the original and the encoded spectral components.
Dynamic – the artist formerly known as DickD

FM Broadcast: 44 kHz or 48 kHz

Reply #26 –
I was only partially making the assumption that radio "is radio" on the basis that it has a bandwidth that's deliberately limited to around 15kHz and relatively much narrower channel separation than a CD source potentially has. The signal is what it is, and using a -V setting to tune the encoder to ignore what the signal isn't and throw bits at what it is may (or may not) produce much better results than you think.

I have no easy way of getting the output of a good quality FM tuner into my soundcard, so I can't really do my own testing to prove or disprove this, but I'm happy to be proven wrong by somebody else's ABX test results.

@Dynamic: I understand that the level of errors is going to go up, but is this a significant factor when the output from a typical FM tuner contains a relatively high level of distortion compared to original source material anyway, as well as a noise floor of only around 60dB at best to hide things under. Won't that help to mask the errors to a degree? If so, then by enough to make the additional errors not matter? Is it ABX time?

Cheers, Slipstreem. 

PS Apologies if I seem like a bit-miser, I just like to see the bits being put to the best use possible before I'm prepared to up the bitrate.

FM Broadcast: 44 kHz or 48 kHz

Reply #27 –
Quote from: Slipstreem on
Is it ABX time?
By public demand. Two clips received using a piece of wire. Station ID jingles, and an announcer's voice.

I regret that I picked this sample because the synthetic instruments are intentionally distorted, making them sound like mp3. Also I have a feeling that the studio has non-transparent lossy compression employed somewhere. The voice sounds very unnatural to me. Still I was amazed how hard it was to spot any differences, considering the avg bitrate. I had to concentrate on the voice and the beginning of the second jingle, where there is a pronounced stereo separation.

- wirepiece2.mp3
LAME 3.97 -V7 --vbr-new (102 kBit/s)
- wirepiece2fft.mp3
LAME 3.97 -V2 -k (173 kBit/s)

Recorded using E-MU 0404 at 48000 Hz using 24-bit resolution. Normalized. Second clip processed with FFT filter in Cool Edit Pro.

FASTER (maybe): http://rapidshare.com/files/133193162/wirepiece2.zip.html
SLOWER: http://j7n.sytes.net/temp/wirepiece2.zip


Code: [Select]
foo_abx 1.3.1 report
foobar2000 v0.9.4.5
2008/07/29 02:23:51

File A: H:\temp\wirepiece2.mp3
File B: H:\temp\wirepiece2fft.mp3

02:23:51 : Test started.
02:25:11 : 01/01  50.0%
02:26:30 : 02/02  25.0%
02:27:50 : 03/03  12.5%
02:29:11 : 04/04  6.3%
02:31:47 : 05/05  3.1%
02:33:39 : 06/06  1.6%
02:34:44 : 07/07  0.8%
02:35:55 : 08/08  0.4%
02:37:16 : 09/09  0.2%
02:38:44 : 10/10  0.1%
02:38:47 : Test finished.

 ---------- 
Total: 10/10 (0.1%)

Quote
Good antennas are great and their value should never be underestimated. A nice multi-element is going to provide a cleaner signal by virtue of having a relatively narrow field of view and a very good front/back ratio when compared to the 360 degree field of view of a straight piece of wire.
Yes, a real anntenna will be able to reject strong unwanted signals. If you live next to a transmitter, without a good antenna you might not be able to receive any remote TXs. Besides a wire is hard to fix in a certain position. I did had reception worthy of recording with a wire. But then I had to relocate some some furniture and could never get the makeshift antenna back in the previous place after that.

I chimed in because I disagree with the statement that a radio recording is not worth a reasonable amount of bits. Lot's of people think that way and I'm forced to download content like this.

FM Broadcast: 44 kHz or 48 kHz

Reply #28 –
I think the suggestion was to use -V2 with a low-pass.  -k is not a low-pass!

...or does the FFT applied with Cool Edit Pro result in no difference compared to applying the filter with Lame?  If this is the case, does -k even have any effect?

EDIT:  After looking at the clips, what happens if you use -V3 --vbr-new --lowpass 14.5?

FM Broadcast: 44 kHz or 48 kHz

Reply #29 –
I used -k because I think there is no point lowpassing again. I don't rely on LAME's polyphase lowpass filter while recording radio due to reasons explained here.

"LAME.exe --lowpass 16.500 -V2" produced a significantly smaller file. While the cutoff was the same, a glance at the spectrum analyzer showed that more low level signal is missing (not that I claim it was audible). My intention was to to prove that -V7 is not transparent, also for radio.

Quote
After looking at the clips, what happens if you use -V3 --vbr-new --lowpass 14.5?

It downsampled to 32000 Sa/s! Is it supposed to be doing that?

FM Broadcast: 44 kHz or 48 kHz

Reply #30 –
I was working from the bitrate angle.  Sorry for implying that you thought -k was a lowpass.  I know you know better.  I don't think the -k should make any difference since there should be no data in the sb21 band after low-passing, but I could easily be wrong.

To be honest, I jumped in not realizing you had pre-filtered the data prior to encoding.  I'm pretty curious about what will happen with the setting I asked about, but since goal is transparency (as you rightfully mentioned) maybe take the low-pass up about a kHz.

Quote from: j7n on
It downsampled to 32000 Sa/s! Is it supposed to be doing that?
That would make the most sense.  I wonder how high you can get before it goes back up to 44.1.

FM Broadcast: 44 kHz or 48 kHz

Reply #31 –
Bitrates.
-V2 -q 0 -k (48000 Sa/s, filtered at about 16.5k) -- 173 kBit/s
-V2 -q 0 --lowpass 16.500 -- 156 kBit/s
-V3 --vbr-new --lowpass 14.500 -- 145 kBit/s
-V7 --vbr-new -- 102 kBit/s


The voice only.
1) -V3 --vbr-new --lowpass 14.5
2) -V2 -k 48000 Hz
Code: [Select]
foo_abx 1.3.1 report
foobar2000 v0.9.4.5
2008/07/29 04:28:22

File A: H:\temp\divosnaktii32.wav
File B: H:\temp\divosnaktiiK.wav

04:28:22 : Test started.
04:30:37 : 01/01  50.0%
04:30:45 : 02/02  25.0%
04:32:53 : 03/03  12.5%
04:33:28 : 04/04  6.3%
04:33:53 : 05/05  3.1%
04:34:33 : 06/06  1.6%
04:34:52 : 07/07  0.8%
04:35:09 : 08/08  0.4%
04:35:19 : 09/09  0.2%
04:35:27 : 10/10  0.1%
04:35:34 : 11/11  0.0%
04:35:48 : 11/12  0.3%
04:36:03 : 12/13  0.2%
04:36:10 : 13/14  0.1%
04:36:17 : Test finished.

 ---------- 
Total: 13/14 (0.1%)

I carefully cut exactly 0:02.130 where the announcer's voice was. I'm most sensitive to unnatural voice, maybe because I'm used to hearing synthetic music. It was hard to notice any difference. I would definitely not complain about such quality. After training for a couple minutes I realized that the resampled version was a bit smeared. The sharp sibilance is actually a kind of distortion, though I'm not sure if smearing it is a 'fix'.

I compared against the very high quality MP3 and not lossless, what I realize now was not the best thing to do. I chose to do that because using lossless never came up as an option, and to make the comparison reproducible by somebody else without downloading 20 MB of data. Feel free to disregard the results if you see fit.

Quote from: greynol on
Sorry for implying that you thought -k was a lowpass.
No need to be sorry.
 

FM Broadcast: 44 kHz or 48 kHz

Reply #32 –
Thanks for the data so far. I'm slowly soaking it up. I think I can understand why it's best to stick to the same -V setting that you'd normally use but to set a low-pass filter at around 15kHz now. You still gain in terms of a small bitrate reduction but without losing any more quality than you would with a full 20kHz source, relatively speaking.

Apologies if going off on a tangent confused anybody.

Cheers, Slipstreem.