Voice Recording Noise problems

SanDisk MP3 players Sansa® Clip | Sansa® Clip+
1

sansafix wrote:
On the clip you have no limit on the voice record time except the flash capacity.
We will consider a future feature to allow adjustment of Voice record quality.  At this point its not part of the product spec.

SansaFix, are you aware of the noise issue with voice recording? When I first bought my 2 GB Clip, a must have feature was the voice recording. I love the rest of the device features, and I definitely appreciate the responsiveness of the Sansa support, but this voice recording feature leaves a lot to be desired.

First the format of WAV is weak. I realize that compressing while recording is processor intensive, but I am not looking for perfection. My old Lexar LDP600 (512MB) could record in 3 qualities, in WMA format, all at 44kHz sampling:

Low - 32kbps, Mono

Middle - 64kbps, Stereo (yes stereo)

High - 96kbps, Stereo 

Second, when recording using the 20a firmware, there was an audible chirp every so often in the recording. After upgrading the firmware to 29a, that issue seems to have gone away. Recordings sound clean when played back, but when examining the files in the Goldwave editor (which I was using to convert to MP3), I noticed a low-level, repetative noise signal throughout the entire recording. This shows up in every recording made with the device.

I have compared waveforms of simultaneous recordings made with both the Clip and my Lexar, and the Lexar recording is absolutely clean (plus the recording is more full-bodied). The Clip recording shows the distinctive noise signal superimposed on the recording. Also the Clip recording seems to not have as much bass and low mid-range in the signal as it sounds more “tinny” when compared to the more natural sounding Lexar recording.

I can send you samples of both to compare, and I can send you a picture of the repetative noise signal on the waveform of the recording if you desire. I really like my Clip, and want to see it improve. Also I am contemplating getting a larger (memory size) Fuse (or Clip) in the future, provided it can also do voice recording - with a cleaner result. 

BTW I originally ordered the 4 GB Clip but immediately cancelled the order when I discovered the display issue. I eventually purchased the 2 GB model because I didn’t see the root cause of the display issue being addressed on the 4 GB Clip (it’s the coating - not the firmware).

Here are links to two files - the original recording test file, and the associated picture (screen capture) of the noise in the file.

This recording was made using the latest firmware. It has the same issues as the 29a firmware.

http://www.4shared.com/file/76697919/1167b5e9/VORC001.html
http://www.4shared.com/file/76697996/49012270/Voice_Recording_waveform.html

Message Edited by CyberRon on 12-29-2008 12:38 PM

2

Ron,

I listened to the sample you posted, and I do not hear any particular audible noise.  You say it’s a “repetitive” noise so I expected a repeating beep, chirp, or whatever; I don’t hear any such noise.  Of course it’s hard to tell because you are talking, there is noise from you handling the device, room noise, etc. 

If you want to make a meaningful sample to listen for really low level noise, I suggest you do it on a really quiet day (or night).  Start the clip, then place it on your bed, and cover it with one or more pillows, blankets on top of that, etc.  (Sounds cheesy, but it works.)  Then stand clear of the bed so you’re not jostling it in any way, and record at least 30 seconds of silence.

I also looked at the image you posted.  I would not really call the top part of the image a “waveform” because of the large solid green area.  There’s no way to tell what that is from looking at it.  A waveform graph should be a continuous line, but not any solid area.  The part that looks solid might represent a continuous high frequency  wave that is too high in frequency to hear.  But there’s no way of telling, from looking at that time scale.  You’d have to zoom in repeatedly to see what it is that looks like solid green.

Since I was curious, I opened your .wav file with CoolEdit.  My screen capture, posted here http://nittany.ws/images/CyberRonSansaClip-sample01.gif looks significantly different.

Note that CoolEdit does not display a solid colored area.  Instead, it shows that the baseline has a positive polarity with an amplitude of about -18dB (relative to 0dB full scale).  That means that there is some DC offset in your unit, which is being accurately captured and recorded into the .wav file.  (And I think that means that the software you used to generate your graph is flawed… it doesn’t really show the waveform, because it fills in with solid green when it shouldn’t.)

The good news is that a DC offset like that is not audible, per se.  You can’t hear DC.  At worst, if you play this back over a DC-coupled amp, at high gain, your speakers might make a “thump” sound at the beginning of the file.  (You can see where the voltage ramps up from infinity (no voltage whatsoever) to the -18dB level, near the beginning of the file.)

The bad news is that a DC offset reduces your available system headroom.  In theory, a 16 bit recording gives you 96dB of dynamic range.  2 to the 16th power = 65536, so the smallest signal you can represent is 1/65536 as big as the largest signal.  Take the log of 1/65536, multiply the log by 20, and you find that’s -96db (approximately represented by the centerline on the graph).  But since your offset is at -18, you do not have the entire 96dB to work with.  So if you’re recording something fairly loud, the Clip will clip sooner than it should.  (Maybe that’s where they got the name!)

Now, depending on where that DC offset comes from, it might mean that the record chain has really good low frequency response.  Or it might just mean that the A-to-D converter is lousy.

At any rate, I’d like to get a Clip, if I can find one cheap enough, and try to open it up and do a little more research.  I see them for $20 on eBay, but I’m holding out for $10, since I might kill it when I dissect it.

Meanwhile, if you want to do a more meaningful noise test, make another recording as I suggested, and post that file for me to look at.  Hopefully this BBS will notify me if you reply to this message of mine; but if I don’t get a notification, I’ll try to remember to check back here in a week or so.

Happy Clipping!

Message Edited by gminpa on 02-12-2009 07:55 PM

3

I’m confused,  I downloaded the wav file,  opened it on Sound forge, it plays with very minimal background noise.   I copied a sample of the noise starting at 00:00:05.162 seconds:

                               Left Channel         Right Channel      
Cursor position (Time)         00:00:00.489         00:00:00.489       
Sample value at cursor (dB)    (end of sample)      (end of sample)    
Minimum sample position (Time) 00:00:00.003         00:00:00.003       
Minimum sample value (dB)      -18.983              -18.983            
Maximum sample position (Time) 00:00:00.130         00:00:00.130       
Maximum sample value (dB)      -17.369              -17.369            
RMS level (dB)                 -17.938              -17.938            
Average value (dB)             -17.940              -17.940            
Zero crossings (Hz)            0.00                 0.00               

It’s pretty well dead silent.  What kind of room / acoustics was it recorded in?   How about how you are positioning the mic vs. other objects.  If you are expecting better recording from the clip,  I don’t think you will get much better.  It’s already excellent.

Message Edited by niko_sama on 02-12-2009 07:21 PM

4

I’m confused,  I downloaded the wav file,  opened it on Sound forge, it plays with very minimal background noise.   I copied a sample of the noise starting at 00:00:05.162 seconds:

                               Left Channel         Right Channel      
Cursor position (Time)         00:00:00.489         00:00:00.489       
Sample value at cursor (dB)    (end of sample)      (end of sample)    
Minimum sample position (Time) 00:00:00.003         00:00:00.003       
Minimum sample value (dB)      -18.983              -18.983            
Maximum sample position (Time) 00:00:00.130         00:00:00.130       
Maximum sample value (dB)      -17.369              -17.369            
RMS level (dB)                 -17.938              -17.938            
Average value (dB)             -17.940              -17.940            
Zero crossings (Hz)            0.00                 0.00               

It’s pretty well dead silent.  What kind of room / acoustics was it recorded in?   How about how you are positioning the mic vs. Other objects.  If you are expecting better recording from the clip,  I don’t think you will get much better.  It’s already excellent.

It’s possible you are hearing an ambient noise from the room around you.  Maybe the clip is attenuating it a bit from its plastic case.  Similar to a sea shell.   But I am not an audio engineer.   I’d repeat a test here, but I have too many noisy things within this room.  windows, computer fans, etc.  

5

@gminpa wrote:

Ron,

 

I listened to the sample you posted, and I do not hear any particular audible noise.  You say it’s a “repetitive” noise so I expected a repeating beep, chirp, or whatever; I don’t hear any such noise.  Of course it’s hard to tell because you are talking, there is noise from you handling the device, room noise, etc. 

 

If you want to make a meaningful sample to listen for really low level noise, I suggest you do it on a really quiet day (or night).  Start the clip, then place it on your bed, and cover it with one or more pillows, blankets on top of that, etc.  (Sounds cheesy, but it works.)  Then stand clear of the bed so you’re not jostling it in any way, and record at least 30 seconds of silence.

 

I also looked at the image you posted.  I would not really call the top part of the image a “waveform” because of the large solid green area.  There’s no way to tell what that is from looking at it.  A waveform graph should be a continuous line, but not any solid area.  The part that looks solid might represent a continuous high frequency  wave that is too high in frequency to hear.  But there’s no way of telling, from looking at that time scale.  You’d have to zoom in repeatedly to see what it is that looks like solid green.

 

Since I was curious, I opened your .wav file with CoolEdit.  My screen capture, posted here http://nittany.ws/images/CyberRonSansaClip-sample01.gif looks significantly different.

 

Note that CoolEdit does not display a solid colored area.  Instead, it shows that the baseline has a positive polarity with an amplitude of about -18dB (relative to 0dB full scale).  That means that there is some DC offset in your unit, which is being accurately captured and recorded into the .wav file.

 

The good news is that a DC offset like that is not audible, per se.  You can’t hear DC.  At worst, if you play this back over a DC-coupled amp, at high gain, your speakers might make a “thump” sound at the beginning of the file.  (You can see where the voltage ramps up from infinity (no voltage whatsoever) to the -18dB level, near the beginning of the file.)

 

Now, depending on where that DC offset comes from, it might mean that the record chain has really good low frequency response.  Or it might just mean that the A-to-D converter is lousy.

Message Edited by gminpa on 02-12-2009 07:55 PM

Thank you your response and analysis. I took a closer look at the original waveform and can see the offset rising from the centerline (0.0 on my Goldwave graph) to a steady DC level. I don’t know why it isn’t rendered that way. I zoomed in to a small section that was relatively flat, and kept doing so until I got a sraight line showing the DC. I then backed off one zoom level and saw the solid mass again. The time increment change is very small. I don’t know why this isn’t rendered properly - something to ask Goldwave.

I will attempt to make the recording as you suggest and see what happens. It looks to me that the offset begins at zero and increases until it reaches the steady-state offset that remains for the remainder of the recording. The ramp-up has a characteristic shape of a leaky capacitor somewhere in the audio processing circuit. It always starts at zero and ramps up to the same steady state value regardless of recording environment. Fortunately Goldwave can easily remove it, and as you said - it is not audible.

 Here is some screen shots of the displays.

Noise Shown as a solid “bar”

One zoom level in - DC offset shown

6

hi All , you can remove noise from almost any voice and sound with Adobe Audtion. and that is give good result almostly. . firstly u must select pure nosie part from your sound without any background from mainly source. then process your sound.

you can give more info via google search.

With my best regards

7

i was recorded voice from class almost 3 H  and my recorded voice have noise . untill first 20 min,noise is weak but at 120 min ,noise is strong. i think that clip is not good for recordnig and i wish that this issue will be solve with new frimware .and adding bittrate setting to sansa and change record format to mp3.

Thanks sansa.