Naked Science Forum
General Discussion & Feedback => Radio Show & Podcast Feedback => Topic started by: Thrash100 on 05/06/2008 12:55:46
-
Hi, I love the show, but sometimes I find it unlistenable because of the amount of siblance on the voices. Short bursts of high pitched noise on certain words. Please tweak the settings on your multiband compressor or introduce a de-esser in post production. I am a sound engineer and have trained my ears to hear these kind of things, so I am probably more concious of it than most people. I would be glad to help if needed.
I also noticed the mp3 podcast could possibly be encoded with more suitable settings. The current mono 48kbps is fine for a talk show in my opinion but the file really should be normalised and resampled to 22khz before encoding. Also, LAME is considered the best mp3 encoder, so if your not using it already you should consider switching to it.
Best Regards,
Kev
-
Hello Kev
the recording comes off a BBC desk and it's hard to get much better hardware. The whole thing, bar some ISDN lines, is normalised to -12dB at broadcast. The noises you describe are artefacts of the compression. I'm not sure why resampling to 22k would make any difference, perhaps you can explain?
Chris
-
I'm guessing that the theory is that by resampling to 22kHz you would then use more "bits" in the compression to code the important lower frequencies accurately, rather than "wasting" bits coding frequencies above 11kHz... in this way you might improve the perceived quality.
Having just watched a spectrum-analysis of a couple of recent TNS podcasts I can confirm that there is an abrupt audio-frequency cut-off at 10.6kHz - so I very much doubt that resampling to 22kHz would make much difference.
Actually, the podcasts are not consistent. Most of the recent podcasts go to 10.6kHz with very little adaptation, whereas a few go to around 16kHz with more "adaptive" bandwidth depending on audio content. Strange...
TNS 08.06.04 goes to c. 16kHz adaptively
TNS 08.06.01 goes to c. 10.6kHz non-adaptively
TNS 08.05.25 goes to c. 10.6kHz non-adaptively
TNS 08.05.19 goes to c. 10.6kHz non-adaptively
TNS 08.05.11 goes to c. 10.6kHz non-adaptively
TNS 08.05.04 goes to c. 10.6kHz non-adaptively
TNS 08.04.27 goes to c. 10.6kHz non-adaptively
TNS 08.04.20 goes to c. 10.6kHz non-adaptively
TNS 08.04.13 goes to c. 16kHz adaptively
TNS 08.03.30 goes to c. 10.6kHz non-adaptively
I do agree the podcasts are a bit sibilant - especially noticeable on Chris' voice ... but I'm not convinced that it's really "distortion" as opposed to just the signal from the microphone. However I'm a physics/electronics guy, not a sound engineer (and have significantly imperfect hearing anyway).
I can confirm that the audio level is normalised and pretty consistent according to the meters on my homebrew audio-analysis software. (Try the SpecAn program from http://www.techmind.org/audio/ )
-
The noises you describe are artefacts of the compression.
Ah, so is there any chance you could please increase the bitrate or supply an alternative high quality version? Dr. Karl and Material World are both encoded at 64kbps.
-
We use 48k to keep the bandwidth down and to minimise download times. The podcast moves about 4 terabytes of data a month and, historically, with those sorts of loads, the extra 20k made a huge difference.
We could, without too much trouble, now increase the bitrate to 64k, but I must say that you are the first person to raise this as an issue; we'll need to discuss it in our team to see whether we think it's currently worth altering it; perhaps we'll include an item in our feedback form (due out soon) regarding the audio quality to see what people think. It's worth bearing in mind that most people who listen do so using cheap headphones from an iPod; I listen back to our show each week, purposefully via the 48k version, on a set of studio monitors and it sounds fine on those.
Thanks, however, for raising this.
And to address the question of the frequency spectrum, the programme's encoded using MP3 Pro, although as a basic MP3. This dumps most frequencies above a certain level because we simple cannot hear them.
Chris
-
Hey Chris, thanks for considering upping the bitrate [:)] I think it'll be worth it!
Regarding the encoder, a convenient way to encode using LAME is by using LAMEdrop which can be found here (http://rarewares.org/mp3-lamedrop.php). It is open source and is the result of endless tweaking by very clever audiophiles with golden ears. It has been shown to be superior in quality compared to other mp3 encoders in blind public listening tests such as this one (http://www.rjamorim.com/test/mp3-128/results.html).
Maybe we could do our own double-blind listening experiment to find out the best encoder/settings to use [?] Hmmmm
-
I have looked for some independent data on the integrity of MP3 encoders; as this graph shows, the FhG encoder (which we use) produces the best compression integrity and is significantly better than LAME at the low bitrates we are using (which is why we're using it!) This graph and the accompanying paragraph are published in an article by Will Ryu (http://arstechnica.com/wankerdesk/1q00/mp3/mp3-3.html)
[ Invalid Attachment ]
"The above graph shows the [Mean Standard Error] MSE (arbitrary units) (on the Y axis) between the original waveform and the encoded signal at various bit rates (x-axis). Note that the Fraunhofer (FhG) encoder has the lowest MSE at 128 kbs by a large margin. FhG continues to have the least error up until a bit rate of 256 kbs, at which point all the encoders essentially have the same MSE. There's a diminishing rate of return as the bit rate increases past 192 kbs -- improvements continue with higher bit rates but relative gains become smaller and smaller."
So FhG is superior to LAME at all except very high bitrates.
And if you look at the article's verdict (http://arstechnica.com/wankerdesk/1q00/mp3/mp3-4.html) you'll see that, at low bitrates, the FhG comes off best, in the author's opinion:
"The envelop please: Low bit rate winner (128 kbs): Fraunhofer. It was no contest. While not CD quality, the FhG encoder was sonically the least offensive. And with its superior measurements, it clearly beat out the other encoders at 128 kbs."
This was also my opinion, which is why we use this encoder.
Chris
-
I too used to use the FhG encoder. It was by far the best quality mp3 encoder but took a lot longer than others like Xing. This was back in the days when I was the proud owner of a super powerful Pentium p233mmx.
The article you have quoted tests LAME v3.61beta which was released 14th Jan 2000. This was even before napster came out! Most people had never heard of mp3, using the internet needed a dial-up modem and the first iPod didn't exist.
I have no idea which encoder is best at 48-64kbps today (and I cant find any modern blind listening tests at low bitrates), but I know that a lot of development on audio compression has taken place in the last 8 years.
-
Sure, but the industry benchmark remains FhG, and we're using their latest protocol, so I'd be surprised if we could improve markedly just by switching codec.
What I will do, however, prompted by your comments, is to investigate some settings tweaks to see if we can achieve slightly better results whilst still minimising the bandwidth and download times.
Cheers
Chris
-
Before doing in-depth comparisons of coders/bitrates, I still think you should begin by testing the hypothesis that it is actually the mp3 encoding step (and not just the original audio) that is causing the sibilance. I'm skeptical!
I would suggest (copyright restrictions permitting!) that you provide Thrash100 (and maybe one or two other people) privately with a minimally-compressed, eg 192kbps, mp3 of a show for comparison. If there's little difference in sibilance, then we can drop discussions of bitrate etc. If there is a difference, then it might be worth looking in to.
That said, I'm quite content with the 48kbps mp3 you're currently providing. I found that the sibilance does become more noticeable (borderline objectionable) at high volume levels, but for normal listening doesn't bother me.
Of course it's always possible that the original audio is okay (with perfectly coded but quite high sibilance levels) but that such levels reveal latent problems in the listeners' audio equipment (amplifiers/earphones etc) and that this could be where *distortion* actually occurs...?
I don't think it matters, but from my real-time spectrum analysis I would confidently say that TNS 08.04.13 and TNS 08.06.04 were mp3-encoded using a different program and/or slightly different compression paramters from the other recent shows!
-
Thanks Techmind. Actually there was a problem last week, which was that one of the channels dropped out intermittently so we had to re-mix in the missing information later; as to the week before, I'm not sure, but I'll ask the crew what they're up to to ensure that we're not compressing compressed data...
Chris
-
I done a test of compressing a vocal sample at 48kbps with the FhG encoder included with Adobe Audition and it didn't create any excessive siblant artifacts.
I also had a listen to the show via the BBC iPlayer and the siblance persists.
Thanks for taking this seriously Chris, even though it doesn't seem to bother anyone else. I wish I could have talked this directly to the engineer when the last couple of Slayer albums were being mastered; he would have gotten an earful lol.
-
We have no control over the iPlayer compression, but I think the raw audio for both iPlayer and Podcast is the same (the digital Record of Transmission). This would suggest that it's the raw audio, and therefore the BBC desk, that creates any problems you're hearing.
Is the issue in the general studio links, or in other sections like Kitchen Science or Meera's on-location packages?
If you find the same issues in other audio (Slayer albums) then I would suspect your audio equipment - are you going through an EQ?
As to the fact that some of the pods compress to 16KHz, while the rest go to 10.6KHz, I have no idea. It's possible that the 16KHz ones were compressed at home, so I'll check my settings there.
-
Hi Ben,
I find the whole show excessively siblant, even the jingles.
The comment about the Slayer albums was refering to the distortion caused by extreme dynamic-range compression used to make a lot of modern music super loud (http://en.wikipedia.org/wiki/Loudness_war). Most people dont notice the distortion and lack of dynamic range, but I cant enjoy listening to some brilliant albums because of it.
-
[:(]
-
You what?