Strange noises when recording to mp3

[lee321987]

   Answering Davide:
 
   Yes, the offset is present on WAV's and MP3's.
   Yes, the offset is occurring on both units, but it's much stronger on the c250.
   I don't know about e2x0 units.

[saratoga]

If you're encoding to mp3 anyway, lame can be told to remove the DC offset using its highpass filter option.  However, since people cannot hear a DC offset, and because DACs should filter them out, it shouldn't impact quality.

[lee321987]

   But it does seem to be the problem (at least with the Rockbox mp3 encoder).  I'm using the exact same version of Rockbox on both players (not even a different "../tools/configure" number), and the only difference I can find is that the c240 produces a much smaller offset (and doesn't produce this noise).   
   I'm not putting down Rockbox's encoder.  I think the size of the offset produced by my DAP may be a good deal over average.

   Anyway Rockbox still doubled the amount of recordings I can store internally by giving me WavPack (the OF only offered WAV), and multiplied MANY times my total capacity by allowing me to record onto the MicroSD.  Oh yeah, AND gave me 22050 Hz sample rate (OF only offered 16000 Hz).  Go Rockbox!

[Multiplex]

But couldn't the offset be causing the noise?

From a PCM value of 0V to whatever the offset is may look like a sharp edge, the encoder probably can't cope with that and noise is the result of it trying to encode it.

Look at a JPG of this web page for example, the sharp transitions will cause visible noise when you zoom in.

[dreamlayers]

I doubt the WAV is totally silent.  There's probably some quiet sound from the microphone and maybe some noise which is like white noise and less noticeable and less objectionable.  One could take a lossless recording, remove the offset, normalize the quiet section and see if there is some noise.

I don't think the edge is the problem, because the the problem is constant and not just at the start and end of the recording.  I think the DC offset itself is a problem.  Like I said earlier, a DC offset even messes up LAME, resulting in bass pulses.  (As someone else said, with LAME it is possible to remove that with the high pass filter.)

[lee321987]

   I did a test, a link to my results is here.  In the zip is a file called "original.wav", this is a short recording of silence made using Rockbox on my c250. 
   Description of the other files:  The files with "RB_encoded" in their name were made on my c250 with Rockbox's "Open With... > mp3_encoder" command.  The files with LAME in their name were made with Audacity using LAME 3.97.  The file "offset_corrected.wav" is "original.wav" with the offset corrected by Audacity, and all mp3's with "offset_corrected" in their name were encoded from this wav.
   
   I now have no doubt that the culprit is my c250's large DC offset.

[Multiplex]

   I now have no doubt that the culprit is my c250's large DC offset.

So I wonder how practical it would be to do a custom build to subtract the offset from the data comming from the ADC...

[Davide-NYC]

If this noise bothers you it means you need to bite the bullet and buy a better target.
The DC offset on a recording @ 22k/16bit from a super crappy mic is not worth fretting over.
Better targets do not exhibit this behavior.
I would suggest buying an iriver h120.
If you need timestamp RTCMod it.
If you need total silence (no HD spinup) put a CFCard in it.



Then again, there might be some hardware guru that can resolve this with a paperclip, some chewing gum and a piece of tinfoil. I'm not that person.

I will mention that if Rockbox keeps getting ported to crappy target with even crappier recording support (which I fully endorse, the more target supported the better) it might be a reasonable idea to try to implement an (optional) auto-DC-offset filter for recordings on targets that have the problem.

[AlexP]

crappy target with even crappier recording support (which I fully endorse, the more target supported the better)

There really is no need for this.  You may not like certain targets, but plenty of people do.

[petur]

there is an MP3 encoder patch in the tracker that claims to fix stuff like DC offset.
Maybe you can try it out?

[Davide-NYC]

crappy target with even crappier recording support (which I fully endorse, the more target supported the better)

There really is no need for this.  You may not like certain targets, but plenty of people do.

Look, shame on  Sandisk for releasing a unit with an ADC that only does 22k and a design flaw that introduces DC offset. This is shoddy R&D and quality control on their part. DC offset with induced noise and 22k max samplerate is a deal-breaker for me and maybe OP would *NOT* have purchased the C2x0 series had they known of these flaws before hand. Caveat emptor!

My comment had nothing to do with Rockbox and everything to do with concluding to folks that are looking at the C2x0 series as a recording target (for portability probably) that the hardware is limited. For recording meetings and voice notes it'll be great. (I'm thinking about buying one as a 'loaner' device for meetings/classes I cannot attend.) But if you care about audio quality you may want to look elsewhere. (H120)

That said, the idea of implementing an optional DC-Offset filter for targets that need it might be a cool idea but it seems like it's trying to correct something that is inherently flawed.

[Llorean]

The target isn't designed as a recorder though. Calling it a "crappy target" because one of its features (a VOICE recorder for use with the internal mic only) doesn't meet your requirements is overdoing it.

The player itself handles MP3 playback quite fine, and it's very tiny and light. It has MicroSD allowing for practically unlimited storage/variety despite its tiny size. And to boot it has FM radio in case you'd like to sample local culture. His point was that calling it a crappy target over one minor feature that wasn't even an intended selling point of the hardware is somewhat ridiculous. It's not sold as a real recorder, it's sold as having voice recording out of the mic, and even natively records to a VERY low quality file. I don't see anything that suggests that they want to convince users that it's a high quality, or even mid-quality, recorder. It's not a recording target. It's just a player that has some recording features.

Improving flawed hardware is kinda what we do here. Or would you argue against us trying to fix the remote ticking problem in the H120?

[lee321987]

  Good news.  I just upgraded to r17687, and now when recording directly to mp3 the noise (that I started this subject for) is gone, AND the DC offset appears to be completely gone too (for mp3s). 
  If I first record a wav, and then (on my player) use "Open with..." > "mp3_encoder", the offset IS present, but still - not the noise!)
  I assume the fix was due to the changes at r17679. 

  Thanks again to the Rockbox team.