Re: [Freetel-codec2] Freetel-codec2 Digest, Vol 63, Issue 5

Thu, 23 Jul 2015 21:07:38 -0700

Hi Erich
excellent reading.

The codec will always perform more consistently when
a- the spectral energy distribution  is ideal.
b- minimal background noise
c- no tones.

Pre processing :
- (a) - some sort of multiband dynamic audio processor (with gate) .
    or  - adaptive filter attempting to shape output spectrum .

 (b) noise reduction algorithm . There are plenty.  some are better with white noise (LMS), some are more suited to no-white noise (spectral subtraction algorithms etc ) which might be more expected in the real world.

(c) tone removal algorithm. there are plenty of those.  speed and artifacts dependent...

that's what is required for success , and contributes to a fair number of the MIPS required over and above the speech codec.

cheers


On 24/07/2015 1:39 PM, Erich Heinzle wrote:
It seems the emergency responders in the US Fire Services looked at background noise and P25 radio codec performance for their safety critical workers soon after its introduction.
 
 
 
FWIW, the Utah site mentions how companding may confound encoding of low level audio in the presence of background noise.
 
Regards,
 
Erich.
 
 
Message: 3
Date: Fri, 24 Jul 2015 09:33:15 +0930
From: David Rowe <[email protected]>
Subject: Re: [Freetel-codec2] microphones
To: [email protected]
Message-ID: <[email protected]>
Content-Type: text/plain; charset=windows-1252; format=flowed

Thanks Jean-Marc,

Just had a chat with some local Hams on our morning "drive time" net and
I have a radio receiver model:

+ a microphone is an antenna with a noise figure.  We assume linearity
for now (the microphone isn't clipping and is linear).

+ To remove some variables lets say we take extacly the same mic and try
it 2cm and 50cm away from the speakers lips.

+ If you take the same microphone and move it away the signal power
drops and (given the same noise figure) SNR must decrease. Adding extra
gain doesn't help the SNR, just like adding gain down the track in a
radio receiver doesn't help the SNR.

+ Over a longer path, different frequency components will be attenuated
by different amounts, as the lips are not equally efficient in radiating
at all frequencies.  Here we diverge from the radio analogy; radio
signals are usually narrow compared to the antenna bandwidth.

+ A microphone further away picks up multi-path reflections from the
room, laptop case, that start to become significant compared to the
direct path.  Summing a delayed version of the original signal will have
an impact on the frequency response and/or add reverb - just like a HF
or VHF radio signal.  These may be really hard to remove.

-/-

Re the Speex denoiser, yes good idea.  I have that in command line unit
test form in codec2-dev (so I can pipe audio files through it) and BTW
we are also using it in the FreeDV GUI program where it's helping with
background noise - thanks Jean Marc :-)

-/-

Gd point on the energy of the low freq speech components - I am looking
into a related problem (see Glen's post) with samples that have much
greater LF than HF energy.  These tend to do poorly, which I think is
due to the LPC (short term spectrum) model struggling with the high
dynamic range. LPC analysis designs a filter to mimimise the energy in
the residual (the signal after filtering), so will "throw poles" at the
high energy parts first.

Our ears, however have a log response to energy, so the codec breaks and
sounds bad (or rather even worse Lol).

Thanks,

David

On 24/07/15 08:16, Jean-Marc Valin wrote:
> Hi David,
>



------------------------------------------------------------------------------



_______________________________________________
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


-- 
- 
Glen English
RF Communications and Electronics Engineer

CORTEX RF
&
Pacific Media Technologies Pty Ltd

ABN 40 075 532 008

PO Box 5231 Lyneham ACT 2602, Australia.
au mobile : +61 (0)418 975077 


------------------------------------------------------------------------------

_______________________________________________
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

Reply via email to