You are not logged in.

## #1 2005-03-26 15:31:23

Michel
Member
From: Belgium
Registered: 2004-07-31
Posts: 286

### recording: 8bit->noise / 16bit->no noise

Heya all,

anyone knows why recording on 8 bit(mono) gives me noise and recording on 16bit(mono)(SE_LE) gives me no noise? I'm using alsa.

When I convert the 16bit noise to 8 bit using sox I also get noise.

I ask this because when I use gnomemeeting with the ms-gsm-codec I get noise. I'm not sure if it uses 8bit-sound.

Is there a ladspa-plugin to denoise an audio-stream ... or maybe some tips on how to do this? I just noticed that I can attach ladspa-plugins to alsa-pcm-devices ... (alsa-ladspa)

Thanks,

Michel

Offline

## #2 2005-03-26 17:39:04

incinerator
Member
From: Edinburgh, Scotland
Registered: 2005-02-15
Posts: 80

### Re: recording: 8bit->noise / 16bit->no noise

With 8 bit digital sound, you will always get audible noise. That is caused by the poor resolution of 8 bit sound. Audio sampling is simply measuring an analogue signal and giving it a discrete value. For 8 bit, you will have to choose between 256 different values you can give that particular sample.

Now, it should be obvious that the digital sample value will almost never resemble the exact analogue signal level. We can assume that the maximum error for each sample will be 1/2 of the level difference between two adjacent discrete values. For an 8 bit signal, the maximum error is 1/512 of the peak signal level. This error is called quantisation error.

The point, this will distort your signal, every sample in your audio waveform will have a quantisation error. However, for each sample that error will be random, its value ranging between zero and 1/512 of the max signal level. "Quantisation noise" will get "added" to your signal, having a "strength" of 1/256 of your original signal you want to sample. Doing some math you will find out that "quantisation noise" alone will have a signal to noise ratio of ca. 48dB, which is well audible for human ears. However, that is only the best case scenario. Other equipment like the analogue playback device you sample from, the a/d converter and playback equipment like your d/a converter, amplifier etc. will generate additional noise.

With 16-bit audio, the signal to noise ratio for quantisation noise will be raised to ca. 96dB. That is not significant for human ears anymore.

From my experience with Amiga equipment the noise should not be to strong, though. With normal music you probably won't hear it except you know what you're "looking" for (or if you do a direct comparison).

Another point is that lossy audio codecs will generate "noise", as well. The cause for this is the very data loss achieved by lossy compression, having effects similiar to quantisation noise. Extremely high compression ratios will cause the converted signal to be noisy. If you encode audio with high compression codecs (like speex or gsm codecs) make sure the encoding programs use frequency filters (sometimes called low- or highpass filters). If they don't you should apply such filters before encoding, using an audio processing tool like audacity or whatever. Decreasing the compression ratio may also help.

See wikipedia for a proper explanation of quantisation.

Cheers,
Dominik

Offline

## #3 2005-03-26 19:50:04

Michel
Member
From: Belgium
Registered: 2004-07-31
Posts: 286

### Re: recording: 8bit->noise / 16bit->no noise

Thank you for the explanation Dominik. It is nice to know this.

Michel

Offline