[Speex-dev] how to fix high freq noise?
jhennigar at pantelintl.com
Fri Jan 9 11:50:56 PST 2009
I just retested the speex audio with speex-svn-14525 and speex-svn-15540
with the original Blackfin makefile configuration settings. The two
tones are still present. The original setting are below.
In order to use these settings with the speex-svn-15540 a minor
modification was required to the preprocess.c module. agc_level is not
declared when fixed point is used.
< #ifndef FIXED_POINT
< #endif /* #ifndef FIXED_POINT */
Jason Hennigar wrote:
> Hi Jean-Marc,
> Thanks for the quick response! I am resending this mail hopefully in
> plain text.
> The following three version have been tried and all of them have the
> high frequency tones.
> I've isolated this down to the Speex encode/decode by sending silence
> (i.e. all zeros) to the encoder and displaying the decoder output in
> addition to recording the audio. The ALSA functionality that I am
> using to receive and transmit audio is not introducing the problem as
> it works correctly when I remove Speex. The problem occurs both on the
> Blackfin 537 Stamp board with the AD1836 audio daughter card and with
> our own 537 hardware.
> The build configuration settings that I am using are as follows:
> Jean-Marc Valin wrote:
>> What version did you use. Some older releases had encoder-side issues
>> with some times, but that should be fixed now.
>> Jason Hennigar a écrit :
>>> Hi Masakazu,
>>> I have reproduced the Speex high frequency noise issue on a Blackfin
>>> 537 by sending 'silence' into the Speex fixed point encoder. By
>>> recording the Speex decoder output and plotting the spectrum using
>>> Audacity the following two frequencies.
>>> 1596Hz at -31.3dB
>>> 3200Hz at -48dB
>>> These two frequencies are present in both Narrow Band and Wide Band
>>> modes. I've tried a number of different configurations and these two
>>> frequencies are always present (quality set to 10 or default, the
>>> complexity set to default, perceptual enhancement turned on or off).
>>> I have not determined how to correct this problem but wanted to
>>> provide additional information.
>>> Speex-dev mailing list
>>> Speex-dev at xiph.org
More information about the Speex-dev