Looking at http://www.di.fm/ I see that the basic streams are:

A. 96k MP3 (44.1kHz, JS)
B. 24k AAC+ (44.1kHz, SBR)
C. 32k WMA (Windows Media Audio 9.1, 32 kHz, stereo 1-pass CBR)

Not seen any tests so far that quantify these against each other.

Premium rate offerings are:
128/192k MP3
32/64k AAC+
64/128k WMA


My initial perception of the non-premium streams is that:
I didn't like the 32k WMA, and that was before seeing that it was downsampled to 32kHz.

96k MP3 - listenable quality on PC speakers (a 5.1 set).

After installing Foobar2000 (no other AAC capable software installed):
24k AAC+ - couldn't spot any difference from 96k MP3


Is AAC+ really that good, or MP3 really that bad? - I assume they ARE aiming for approximately equal quality on each stream.

The other thing, these all appear to be CBR (or perhaps, very firmly enforced ABR, as the AAC+ fluctuates 23-25), which is supposed to be inferior to VBR.

The first thing i can tell you is that since some time ago (i can't tell since when, because i sort of stopped listening to those stations), is that 96Kbps mp3 stream is *better* than what it used to be, and comparable to a 128Kbps stream (they upped the lowpass, and there don't seem to be as many artifacts as before).

The second thing is that the 24k aac+ stream is good, but they are using aac+v1, not v2 ( i.e. no PS stereo ).
The lowpass is the same than the 96kbp stream, and sounds quite similarly, except maybe more grainy.

They sound quite well. Artifacts can still be heard depending on the music (pre-echo, flanging ...)

Overall, i'd applaud them for giving quality at such low bitrates



[Edit : changed "noisy" by "grainy"]

I remember when their 128kbps stream was free to listen to

When they first switched to 96kbps mp3, the lowpass was around 12-13khz (guess, I didn't have the tools to measure it at the time). Now it is at 15khz. LAME 3.94 and higher lowpass at 15khz at 96kbps 44.1khz.. so they didn't 'up' the lowpass, it is the default. (they just aren't resampling to 32khz).

Last I checked, their 24kbps aacplus stream was with parametric stereo. I haven't listened in quite some time, but I can't think of any reason why they would change that.

They are not aiming for comparable quality on each stream. The 24kbps stream is for dialup listeners (i remember when that was 24kbps mono mp3 @22050hz), they just use aacplus now for improved quality (*subjective).

For some reason people like to stream in CBR instead of ABR.. something I can't really understand, especially when 10-30 seconds is buffered anyways

I tried 24 kbps AAC+ v2 (with parametric stereo) encoded by Nero. The result was pretty good. However, when I compared it with the original (FLAC), it was quite easy to hear that there was a difference. The AAC didn't sound bad, just "different", especially higher frequencies (probably because of SBR).

I don't know which encoders they use for MP3, but I'd prefer 96 k LAME instead of that.


Even ~256 k VBR Ogg Vorbis works fine for me

..and it seems they won't tell
http://forums.di.fm/showthread.php?t=94247...ghlight=encoder

We can only guess what they are using. Anyone care to speculate? (please provide reasoning)

Oh, and the lowpass on the 96kbps mp3 stream used to be ~11.3khz or so.

Mr. Question Man identified it only as FhG... I took a look at it with Goldwave, hi cut off seems to be about 16K.

And, yes, I did a true stream capture not a record (re-encode of the decoded stream).




Mr. Question Man identified it only as FhG... I took a look at it with Goldwave, hi cut off seems to be about 16K.

And, yes, I did a true stream capture not a record (re-encode of the decoded stream).

It will be shown as fhg in mr questionman because there are no extra headers present to identify it otherwise (same thing happens with a stream encoded by lame).. so that doesn't tell us much.

Last time I looked, I thought the lowpass of the 96kbps stream was 15khz, maybe they upped it again to 16khz. I'll probably use some-some-software-I-shouldn't-mention-here and see what I can come up with. (randomly picking an mp3 encoder and saying that it is what they are using would probably be just as effective/accurate as any guess I can make )

I took a closer look at di.fm's 96kbps mp3 stream today... the lowpass is still at 15khz, and mp3guessenc thinks they might be using lame.
So if it is lame, that leaves:
3.93.1 or earlier,
3.94b through 3.98,
or 4.0a
but which one I'd like to guess 3.93.1 or earlier with --nspsytune... but I'm not entirely convinced. It doesn't appear to be an fhg encoder, as there are no frames with intensity stereo. The only other possibility I can think of might be xing/helix.

I must find out what they are using..

In my findings, the encoder and setting that is VERY close to the quality of their mp3 stream, is LAME 3.93.1 with this commandline: --alt-preset cbr 96 -h --resample 44.1 (yes, it uses NSpsytune)

J.M.

Sorry to rehash such an old topic, but this seems to be one of the more appropriate forums of discussion full of audiophiles (and wannabe audiophiles like myself; i enjoy audio but can't afford "pro" equipment...) and a similar topic existed already--albeit 3 years old now!

I have recently started listening more and more to DI.fm's streams as I just bought a 3g cellphone capable of streaming MP3 content with a 3rd party app. I can't listen via the corporate LAN as everything streaming audio/video is WebSense'd into compliance and blocked. 8 megabits of bandwidth ontop of 2 I assume is reserved for VoIP (leeching large downloads doesn't affect calls at all) and I can't even use a paltry 64k for an AAC+ stream!

So I subscribed to DI, and begged the app creator for a test version of their program that allowed HTTP Auth.

At first, I thought I heard some scratchiness in the songs but attributed it to bugs in the new hardware and software I was using. This is a T-mobile G1, running Android with the StreamFurious application. Upon further listening on the Future Synthpop channel and comparing to music I have purchased from eMusic, both at 192kbit MP3 encoding, there is a HUGE difference in some songs. This has been independently confirmed on a MacBook 2,1 with iTunes, a linux box running mplayer on an emu10k1 Audigy2 along with my G1. All experience the issue, and ontop of this, all stream formats (mp3, aac, high or low bitrate) are affected.

It sounds to me like a 'nearest neighbor' versus linear or sinc samplerate conversion, possibly to/from 44.1/48khz for processing, and back.

Constant tonal sounds such as a higher pitched string sound or a womans vocal have marked harmonic distortion and scratchy sound.

Can anyone confirm that I am indeed hearing the artifacts of sample conversion, and if so, can we try and get a large thread going on DI's own forums to hopefully effect some change?

I am wondering, what is the best way to visually 'prove' harmonic distortion? My instinct says a voiceprint spectrum analysis but I am unsure of the best app to do this. I have tried with Winamp's "sexy scrolling voiceprint" plugin (part of the tiny fullscreen default vis plugins) but this doesn't seem to produce a result thats visual..

I will try and record a few samples compared between my purchased 192k mp3s, and DI's 192k mp3 stream of certain songs. The best example l(worst sounding) song as of yet has been Covenant's "Monochrome" from the Future Synthpop channel. This particular song starts off with single note polyphony string sounds(is this the right term for a single note at once?) that have horrid distortion. I believe that this particular song is also available from eMusic, so I will purchase the song/album and keep listening with a hair trigger on wget to start stream-saving.


Anyone have more input?