advertisement


Radio 3 broadcasting in FLAC

I've now made some lengthy captures of the flac stream. If anyone else is trying this the following may be of interest.

I recorded the evening concerts on the 10th and 11th. During both ffmpeg did throw some error reports relating to TLS and HTTPS/DASH. In short, the TLS ones probably have no effect. but the HTTPS/DASH ones here means a circa-4sec block of audio was missed when the errors were thrown.

Determined this by doing comparisions with the 320k aac version. I can also hear the abrupt jumps in the audio.

My connection has in the past seemed prone to iplayer stream 'pauses' in the evenings, so this may not be something others usually get.

I've also managed to time align copies for the 10th and do a sample by sample 'diff'. This is useful as the result exposes what changes the 320k aac imposes. In places if I play the 'diff' file the results are quite audible. Which does, indeed, support the idea that flac carries details the aac loses or alters. But I'll need to do more to quantify this, etc.
 
Rather than run for 5 months the trial is only going to run for another 20 days. Apparently it's due to the rights agreement in place.
 
This may interest people.

http://jcgl.orpheusweb.co.uk/temp/initialflacresults.png

Its an initial set of results based on a period of about 35 mins from the evening of 10th April. The top graph shows the left/right audio levels in the flac stream. The lower graph comes from subtracting the flac waveforms from the ones from the equivalent 320 aac stream. Having time aligned them to a sample. (The blue and red lines correspond to the stereo channels.)

Note that the lower ("below") plots show how much lower than the *signal* the difference is at any point. So, for example, if the signal level at a given time is -20dBFS and the 'below' is -40dB, then the difference between flac and aac is at the -60dBFS level.

Both plots show the peak level for a series of 0.1 sec blocks of samples. So it probably over-reads compared with a time average. But is shows that the peak difference in sample values between the flac and aac versions tends to run at about 40dB below the actual signal level.

If we assume (not yet confirmed by actual data) that the flac version *is* a sample-accurate version of the source this gives an idea of how much the aac waveform differs from that in detail. But for an average it is likely to be 5 - 10 dB better. However, that's a guess at this point. Nor do I know how typical the above section was as yet.
 
gustavm said:
Anyone know if its possible to try this on an LMS device?
Click to expand...

Afraid I can't say. I'm deliberately using a method that lets me capture the stream without it going though things like browsers, OS audio 'mixers' etc. So I can only point people to the BBC webpages relating to the test. (Which, note, will only run for a few more weeks, alas.)
 
I had a look at a capture, and although it's in 24-bit (nominally, not sure how much of this is noise) it doesn't look like 48 or 44.1, but more like 40 or so:

audacity.jpg


Radio_3_flac01.png
 
Cambridge Audio had suggested some of their streamer users may want to be part of a test group for the R3 flac stream. I politely suggested they actually set about creating software that is usable before they tried anything fancy with streaming flac beta tests!
 
lennyw said:
I had a look at a capture, and although it's in 24-bit (nominally, not sure how much of this is noise) it doesn't look like 48 or 44.1, but more like 40 or so:

audacity.jpg


Radio_3_flac01.png
Click to expand...

The flac stream is 48k/16. But the BBC may decide to trim the top end. However what you're getting if you have the clean flac stream is a straight 48k/16 flac version of the 48k/24 feed the BBC use for internal distribution and the iplayer.

Note that unless you've checked and confirmed otherwise, what you get using a browser, etc, may have been reprocessed in your receiving system.

I'll do some spectra on my captures.
 
This may help clarify:

http://jcgl.orpheusweb.co.uk/temp/spectraflacvaac.png

The image shows two sets of spectra. These are taken from sample-aligned aac and flac stream versions. i.e. the same part of the broadcast. One is from the aac 320k, the other from the flac trial.

You can see that the BBC do tend to roll away as shown. You can also see differences due to the way the aac tries to 'cherry pick' what to convey and what to discard.

The BBC took the decision long ago that it was impractical to try and lock *everything* at the BBC to one central 48k clock. So audio may well end up going though at least one 48k -> 48k asynch reclocking convertors along the way to the central feed that is given to iplayer. if people look on my website they can see the 'Time Lords' page where I show the results of this on a Prom of a few years ago. You can see the clock drifts between different outputs, etc.
 
saturn9 said:
My LMS gives 2 options for R3:
BBC Radio Three (mp3 128kbps)
BBC Radio Three (aac-lc 331.25kbps)
URL for the second option is Link
Hope this helps
Click to expand...

This can be done through LMS using the Dash beta plugin. I'm playing Bbc Radio 3 FLAC now through my Pi3 and IQaudio Dac+. Sounds good.
 
Now have the latest Triode/bpa iPlayer plugin in my LMS, Radio3 FLAC is listed as a separate station, and it sounds superb. My MDAC says it's 16bit/18KHz.
 
JemHayward said:
Now have the latest Triode/bpa iPlayer plugin in my LMS, Radio3 FLAC is listed as a separate station, and it sounds superb. My MDAC says it's 16bit/18KHz.
Click to expand...

Agreed. The sound is less uptight. I particularly notice this in the spoken voice where the hard edge has lessened.Hope they decide to maintain it.
 
You must log in or register to reply here.


advertisement


Back
Top