M1001 noise

[Mitch]

So if the noise really isn't there on 48Khz files, I guess you could re-encode at 48Khz? Obvoiusly you get a (small) quality loss if re-encoding mp3's or spend a huge amount of disk space if you have your music in a lossless format and want to double up to store both 48Khz and 44.1Khz copies.

Another option might be to get your media server to re-encode to 48Khz on the fly. I think mt-daapd can be setup to do this (not sure about wmc2.0). Maybe Ron Peddle could build this into the new Firefly server as an option?

Not an ideal solution, but does work around the problem for now!
Certainly seems like the sample rate converter is at fault.

[grommet]

If someone has a S/PDIF input on the PC, it would be interesting to capture the SoundBridge M1001's digital output and see exactly what is happening to the audio stream. Based on the postings here, that's all the evidience you'd need to prove something is "wrong."

[SoundGeek]

I don't think the data supports any conclusions as of yet. Given the very small handful of people that have responded to this thread hearing a similar thing, and the fact that the Roku guys can't duplicate the problem, I'm hopeful that this will be able to be resolved through the support process or something.

We do know that there have been postings on this thread by people with M1001s (including the Roku crew) that don't have a noise. I don't claim to have "golden ears" or anything close (quite the opposite--I can't hear cable differences that some of my audiophile friends can hear), so I find that encouraging.

Patiently waiting for guidance...

[valdemar]

I can confirm that the noise is NOT present with 48Khz files.

Can anyone suggest a batch conversion workflow for Flac files (44,1 -> 48Khz) without loss of tag information?

Valdemar

[tps]

I have a Soundbridge M1001 and am using it with the DAC in a Sony MDS-JE520 minidisc deck. My causual observation is that the meters do not drop to infinity at any point while the M1001 is playing. I haven't tried playing a "zero" file yet, but I guess I could create one with Cooledit and give it a try. I guess another test would be 44.1 KHz dither with no audio. I'll have to see if I can find such a file.

[SoundGeek]

Two comments:

1) Valdemar, you probably know that CDs are encoded at 44.1 kHz - that's the standard that was set, so if you're going to upsample to 48 kHz at the file level understand that you're just hoping to use a "better" upsampling algorithm than what appears to be used on the Roku. I recommend testing with a high degree of sensitivity before "batch upsampling" your entire library, or you may be no better off. Or actually worse off if you end up with noise at the source - which would be impossible to eliminate. At a minimum I'd be sure to make a backup first!

2) tps, from what I can tell, the Roku upsampling algorithm mutes its output when it sees "digital zeros" per my previous post about testing with the Morrisette CD. Do try the dither and let us know what you find out.

Another great test track to hear the noise is Jeff Buckley's rendition of "Hallelujah" from the Grace CD, as it has a lot of very quiet passages. The noise is clearly audible though much of the song.

[tps]

OK. I have some interesting test results. I made 4 WAV files:

1. 48zero.wav - 10 seconds silence generated by Cooledit 2000 48K/16/stereo. This played fine one the Soundbridge M1001 through iTunes on Windows, with the Soundbridge's S/PDIF output connected to my Sony MDS-JE520.
2. 48dither.wav - 10 seconds silence 48K/32/stereo dithered to 48K/16/stereo with LSB triangular dither. This played fine on the Soundbridge.
3. 44zero.wav - 10 seconds silence 44.1K/16/stereo. This also played fine on the Soundbridge.
4. 44dither.wav - 10 seconds silence 44.1K/32/stereo dithered to 44.1K/16/stereo. THIS EXHIBITED BROADBAND NOISE AT APPROX -50 dBFS, as indicated by the level meters on the MDS-JE520.

So, I'm guessing that somehow the dither is causing a problem. I'm also guessing the the triangular LSB dither generated by Cooledit is likely correlated to the sample frequency. So I guess my next test should be to try to see if I can generate uncorrelated dither.

Edit - my understanding of dither is sketchy. But with the dither depth set to one, ony the LSB should be affected, so in the worst case, the dither noise added to perfect digital silence should be a maximum of - 90 dBFS, leaving 15 bits of the silence. In fact, dither is supposed to give you MORE, not less S/N, by randomizing the least bit quantization. In any event, I should not be seeing anywhere near the noise I'm seeing from 44dither.wav...

Tim

[SoundGeek]

Wow, Tim, that's great stuff. -50 dB sounds about right; the noise is drowned out most of the time, but clearly audible in quiet passages and as songs fade out.

What does 44dither.wav sound like? Can you describe it for us?

[tps]

Listen for yourself. http://proxy1.fielddiagnostics.com/~tps/test.zip

[mundragor]

Hi there,
i got the same Noise on my Pinnacle Soundbridge (Germany).

I riped my CD's with CDex.
CDex have a Setting for the Output Samplerate:

with this Setting, there is no Noise!
But i don't wan't to reEncode my whole CD's.

Hope on a Fix with Firmwareupdate...

[valdemar]

SoundGeek, you are right. Upsampling should happen at time of playback. Then you can use the best algorithm available at that time. - It's just that I could live with a non-optimal upsampling if I could just get rid of the noise.

But now thanks to tps we have what seems to be as close as we can get to a smoking gun. Well done Tim!

Folks at Roku, do you still not hear the noise cranking up the volume while listening to 44dither.wav on a good system?

Valdemar

[SoundGeek]

Well, this is very interesting. I've downloaded and listened to all four of tps' files.

48zero, 48dither, and 44zero are all "silent" when played through my M1001. I put silent in quotes because they may not actually be silent, but I don't hear anything so I'm calling them silent.

I'm hearing two separate things in 44dither:

1) Something that sounds like white noise, which is similar to the sound you get on an FM radio between stations. I have NOT noticed this sound when listening to my music library. This may be the "rumble" that Kim was hearing.

2) A buzzing sound. This sounds like the noise that I've been hearing all along.

Now I'm in over my head from a technical standpoint, as I don't really understand dither, though Wikipedia makes a good effort at explaining it here: http://en.wikipedia.org/wiki/Dither.

My guess is that the white noise I'm hearing is the triangular dither tps used in his samples being made apparent by the upsampling in the Roku...as tps says it must be correlated somehow to the sample frequency...maybe an uncorrelated dither would be quiet. Since I'm not hearing this noise with my music, I'm less concerned about it.

The buzzing, however, concerns me greatly since it appears in all my music.

Roku?

[valdemar]

SoundGeek, it makes a lot of sense to distinguish between the "white noise" and the "buzzing", but I am concerned with both of them.

The "buzzing" is disturbing and the "white noise" hides important low level details in the music, the whole S/N ratio is simply reduced.

Valdemar

[tps]

Someone who has an M500, old M1000 or M2000 needs to try the 44dither.wav file. It should be silent on the those Soundbridge models, since they don't upsample. That, to me, would point a fairly strong finger toward the M1001's upsample algorithm. 44dither and 48dither were made the same way, except for the sample rate, so in theory, they should sound identical. I don't think the broadband noise from the output of the M1001 is the dither signal itself; if it were, it should be audible on 48dither also.

In the meantime, I've had success with upsampling the wav files to 48K using Cooledit, before they go to Quicktime for AAC coding. I trust Cooledit's upsampling algorithm similar to the trust that is put in EAC; it "just works" and provides great quality. MAYBE upsampling doesn't matter that much, but I saw a table at Apple showing that the output sample rate of the AAC decoder is corellated to the input sample rate of the encoder, so maybe Quicktime does the SRC on-the-fly if necessary; I'm sure that Cooledit's SRC is superior to on-the-fly algorithms.

[valdemar]

I have tried out PPHS and SSRC resamplers via Foobar player. SSRC seems to be the most loyal to the original material, actually with no discernible difference to me - for what that is worth.

Valdemar