Radio 3 being broadcast in FLAC

[johnb]

Purely as a test, I captured the AAC and FLAC of tonight's concert and have done a very quick comparison. There is absolutely nothing like the marked difference in Byn's example, though of course that might be influenced by the music being played. (Because of the way I grabbed the files, neither the AAC nor the FLAC would have been subject to the vagaries of Windows processing.)

If I have time tomorrow I will post a link to an example similar to Bryn's. (I was intending to do some work on the central heating system pipework - but as it means the heating will probably be off all day and as the weather forecast isn't exactly encouraging, I might well delay the job until the cold snap passes.)

[Bryn]

The first section is stereo, the second is mono - it must have been mixed down somewhere in the process.

Oops! More haste, less speed. I rarely use Audacity and got one of the settings wrong, resulting in the damned thing recording two identical channels. I think I now have it sorted. Curiously, I got the same problem with Sound Forge Pro 11, plus the level was way too low when capturing the streamed audio. I got better results taking a USB to SP/DIF route to the Alesis Masterlink 9600, but it's such a faff burning from that to AIFF files onto CD-R and then copying the files back to computer. Roll on the delivery of the Tascam DR-100 MK3, hopefully on Wednesday.

[Dave2002]

Here they are: https://we.tl/w9vzYJFuhM for 7 days

I had to find a Flac to Alac converter which works. First I tried MovAvi (which came up as a recommended converter .... !!!) - but that doesn't actually do lossless conversion as is obvious when looking at the output file sizes. Then I found (again) MediaHuman converter (curious title) and converted the files - http://www.mediahuman.com/audio-converter/

This seemed to work OK - though I still have reservations about the conversion. To check that it works correctly I'd have to do a reverse conversion back to Flac and show that the files are identical - which they might not be anyway for other reasons.

The difference file is phasey, as one might expect. This does show that there are differences. Perhaps one could try mixing that back in in various ways to see if one actually notices - though that would also be a somewhat dubious process.

I think there is a difference between the Flac version and the Aac - but it's nothing like as dramatic as in Bryn's example, which does seem to have gone wrong.

Given the uncertainties over whether the lossless conversion (Flac to Alac) has worked correctly this is still a somewhat open question. My hunch is still that the Flac version is better (it ought to be - if the flac process works correctly) but it's difficult to make a definitive assessment when there may be other things going on in the conversion/evaluation process - which possibly invalidate the outcomes.

As suggested, the quality of the source material may also make significant differences - not all tests will show up all the features.

I may have to revisit this using different Flac to Alac converters. The subtleties of the processing are such that unless there is something very obviously wrong it opens up questions about playback anyway. How does one know that the decoding and playback is being done "correctly"? - and this could explain (or not) apparently perceived differences between different software playback systems.

[Dave2002]

I tried converting OT's FLAC files to WAV using XLD - hopefully this is transparent on the Mac.
These files should play on most systems. Now nearly 400 Mbyte download.

The files were converted using XLD and hopefully sufficiently carefully that they are now the correct files - https://we.tl/YhsPOuJTIu

They should be here for about 7 days. The zip file contains FLAC, ALAC, AIFF and WAV versions inside a folder structure.
They are put up for experimental purposes only.

If anyone only wants one variant, and doesn't want to do the download, please PM me.

[johnb]

Oops! More haste, less speed. I rarely use Audacity and got one of the settings wrong, resulting in the damned thing recording two identical channels. I think I now have it sorted. Curiously, I got the same problem with Sound Forge Pro 11, plus the level was way too low when capturing the streamed audio. I got better results taking a USB to SP/DIF route to the Alesis Masterlink 9600, but it's such a faff burning from that to AIFF files onto CD-R and then copying the files back to computer. Roll on the delivery of the Tascam DR-100 MK3, hopefully on Wednesday.

Is there any reason that you preferred using the Audiacity/Sound Forge method (which picks up the data from somewhere in the Windows processing chain) to following up the "nudge" that Dave previously provided a link to?

Curiously, I got the same problem with Sound Forge Pro 11, plus the level was way too low when capturing the streamed audio.

In my test the levels of the AAC and FLAC were very, very close indeed.

When recording with Audacity or Sound Forge the data might well be picked up after being processed by the Windows Mixer - so it is worth checking the effect of setting the Windows Mixer levels at 100%.

[Dave2002]

In OT's samples the difference file clearly has source related content plus noise. Boosted up by 26 dB it is perhaps comparable in level with the actual level - though that is subject to error.

We could write that as D = alpha *S + N where S is the source, N is the noise, and alpha is a small constant.

If everything were accurate, we could probably eliminate the S component almost completely, to find out what the noise N is like, but given that the D component is already 26 dB down, that probably would be rather difficult to do sensibly. alpha is about 0.05 in this case - but this is only approximate, based on 26 dB. Given an estimate of S (what it "really" is) based on the samples, then adding (or maybe subtracting) an approx 26 dB attenuated version of the FLAC version to D should get closer to N - but it will be rather hit and miss.

A curiosity for anyone who wants to try!

I am curious as to how OT claims that he lined up the Flac and the AAC versions to within 1 sample at 98kHz! I find that when playing with video and trying to get audio and video lined up that usually getting the audio within 1 video frame is good enough - though sometimes I can spot problems. That suggests that getting the samples lined up to within 0.01 seconds would normally be good enough, and it could be very hard to detect samples out of alignment by less than 1 millisecond. That's still around 10 of OT's samples!

[johnb]

Dave, when I compared the AAC (decoded to WAV) and the FLAC (both 16/48) in Sound Forge it was surprisingly easy to line up the samples by using clearly defined transient peaks. In fact The samples matched up so clearly that I worried I was mixing up the files.

[jayne lee wilson]

Well, I hope others here revelled in that Bruckner 4 with the LSO & Xavier-Roth as much as I did tonight!

A performance full of interpretative interest with light transparent textures & keenly-profiled rhythms giving onto superbly weighty climaxes underpinned with terrific timpani playing. Xavier-Roth showing - as Knappertsbusch and Furtwängler knew so well - that a rubato-rich shaping of the finale (1878-80 here) does it no harm at all, may indeed be key to a successful interpretation.... hearing it played like it was tonight, somehow both wilful and intuitive, one's doubts about the movement simply fall away.

On the lossless stream, the sound was just magnificent. Smooth, spacious, full and detailed, with remarkable low-level resolution of some extremely quiet string playing just as breathtaking as the thunderous, expansive climaxes. And all so easy-on-the-ears. Did the Barbican ever sound as wonderful as this over the air? Maybe back in the 1990s somewhere on flywheel-tuned FM...
***
I'd forgotten what a lovely slow movement the Bartok Viola Concerto has - and Antoine Tamestit's Viola was effortlessly revealed in all its husky, smoky beauty.
Oh, it will be hard to return to lossy codecs after treats like this - roll on Friday (terrific Bridgewater Gubaidulina/VW program) then a coupla months of patience and anticipation...

[OldTechie]

I am curious as to how OT claims that he lined up the Flac and the AAC versions to within 1 sample at 98kHz!

I got it about right by using the zero crossing on a large transient. Then I trimmed it to get the best cancellation on the difference. I just watched the peak level over the first few seconds with one sample trimmed off from the start of one or other track. I was quite surprised that the difference from one sample change made about 2dB difference in the cancellation as measured by the peak level of the difference audio. It was one sample away from my initial starting point.

And, as usual, Jayne is absolutely right about tonight's concert.

[Dave2002]

I got it about right by using the zero crossing on a large transient. Then I trimmed it to get the best cancellation on the difference. I just watched the peak level over the first few seconds with one sample trimmed off from the start of one or other track. I was quite surprised that the difference from one sample change made about 2dB difference in the cancellation as measured by the peak level of the difference audio. It was one sample away from my initial starting point.

Thanks for the explanation, which makes sense. I note your surprise!

What software were you using to do this? It seems to me that you have more powerful software tools than I have installed.
I have in the past manually lined up audio tracks in video editing software so that there was no perceptible echo, and some video tools can do a form of snapping to line up tracks, and some software can (it is claimed) even line up a whole bunch of tracks automatically - though I think sometimes that doesn't work too well. I was just intrigued that you were working with such fine resolutions, and also I'm somewhat amazed that even a one sample shift made a measurable difference.

[OldTechie]

What software were you using to do this?

Just Audacity. I used its WASAPI connection to grab from the input to my DAC. I had the two recordings on two tracks and just chopped content off the start of one track until it lined up with the other. You can zoom in to the waveform for finer and finer resolution as you get closer to getting it right,

[Bryn]

It's a good job I have hair too short to get a grip on currently, otherwise I might be pulling it out. Even after replicating on a secondary laptop the Audacity and Windows settings which gave me good stereo from the FLAC stream on my main laptop I was still getting two identical mono channels on that secondary computer. Then I noticed I was running an old version (2.0.5) of Audacity on the secondary m/c, so I uninstalled and updated to 2.1.3. Still no joy. However, when I stopped the Taster and then restarted it, suddenly I had stereo! Such things are sent to try us.

The Tascam DR-100 mk3 which I ordered on Monday arrived on time on Wednesday, and tonight, having charged it up and fitted an SDHC card which works (the Tascam does not appear to recognise micro-SD type cards in full size adaptors), and having downloaded the Reference Manual (in the box they only supply a very basic user manual which does not cover recording via the SP/DIF coax input), I have started, during the interval, to record R3iC via USB to Toslink and on to a Toslink to SP/DIF coax adaptor and then on to the Tascam. I should thus now be able to copy and paste bits and pieces from iPlayer and two alternative captures of the FLAC sources.

Tomorrow a new USB to SP/DIF adaptor is due to be delivered. That should give somewhat better results then the USB>Toslink>SP/DIF coax kludge I am using tonight. Watch this space.

[Beresford]

Oh, it will be hard to return to lossy codecs after treats like this - roll on Friday (terrific Bridgewater Gubaidulina/VW program) then a coupla months of patience and anticipation...

Well Friday has rolled on, and wasn't it good? There was a lot of complex quiet orchestral music in the Gubaidulina, one of my all time favourites, and the "Concert Sound" meant I could hear better Sofia's intensity and passion. And Guzman was really up for it. Did I hear him say "the best violin concerto of the 20th century"?

[Dave2002]

Just Audacity. I used its WASAPI connection to grab from the input to my DAC. I had the two recordings on two tracks and just chopped content off the start of one track until it lined up with the other. You can zoom in to the waveform for finer and finer resolution as you get closer to getting it right,

That's interesting.

I did some maths work on the differences - maybe I'll PM you with those for you to check and or comment on. Probably too boring for everyone else round here. In numbers terms there's probably very little difference between the aac signal and the FLAC one, but then human hearing can detect very small differences. Most of us can hear hiss noise on FM or cassettes (remember them) which is down at 30-40dB - ratios of 1:1000 - 1:10000 I think, so in percentage terms very audible noise might only be a near minute fractional percentage of the wanted signal.

Your difference signal seems to have (by inspection - listening) signal correlated sounds - presumably at around -26dB, plus some odd swishing sounds. The signal correlated sounds could simply be attenuated versions of the original, or perhaps some linear combination of phase shifted copies - which if mixed back in would probablyl sound fairly benign to most ears. I haven't been able to establish a definite level for the swishing sounds - probably around 40-60 dB below the wanted signal. When all mixed in, many people may not be able to hear those.

I think the FLAC steam is better, but I can accept that many might not be able to tell the difference.

[Stunsworth]

Recording the output on a Mac is very simple if the free Soundflower extension is installed. Select Soundflower as the output in Audio Midi Setup, and as the input in Audacity. You won't hear anything until you hit the record button, but once that is clicked on you'll hear the stream and it will be recorded.

Apologies if this has already been mentioned.