Re: [Freetel-codec2] microphones

[David R. <da...@ro...>]

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,
>
> Seems to me (this should shock anyone here) like the two main
> differences you'd get between different mics are:
> 1) Frequency response
> 2) Noise
> I've no idea which of the two would be responsible for the bad behaviour
> when coded with codec2. If it's the frequency response, then it *could*
> be due to a mismatch between the response and what the LSP codebooks was
> trained with. It could also be about the first low-frequency pitch
> harmonics that tend to have a lot of power. In either case, the obvious
> solution is to attempt some sort of equalization on the input. The
> tricky thing is of course to properly adapt that equalization.
>
> If the problem is noise, then it may be possible to add a denoiser on
> the input and see if it helps. There's one in Speex. It's not that good,
> but it may still be useful for a quick experiment.
>
> Cheers,
>
> 	Jean-Marc
>
>
> On 07/23/2015 06:02 PM, David Rowe wrote:
>> Hello Lists,
>>
>> Codec 2 has some sensitivity to speech from different sources.  I'm
>> looking into this.
>>
>> I have attached two samples, one from my laptops internal microphone,
>> one from a headset.  The ext sample sounds better when coded by Codec 2.
>>
>> 1/ Can some one tell me why internal microphones sound worse?
>>
>> 2/ Is there anything that can be done to make the internal microphone
>> signal "OK" - or is information lost for ever?
>>
>> I figure there must be a lot of experience on these lists with different
>> microphones.  Traditionally, we hold PTT mics close to our mouths.  Must
>> be a good reason for this.
>>
>> Thanks,
>>
>> David
>>
>> On 23/07/15 18:55, Mark Jessop wrote:
>>> Onno: We've actually done something similar on the local broadcast
>>> (after the national broadcast).
>>>
>>> I second David's comments above about using the command line modems,
>>> will be far easier.
>>>
>>> One thing to note is the end result is going to depend *heavily* on how
>>> the broadcast gets to air. I know around here that the 'originating'
>>> source of the broadcast (VK5RAD) has some pretty bad audio issues, and
>>> some of the other broadcasts (i.e. 40m) are re-broadcasts of that one,
>>> compounding the issue.
>>> The end result can sound pretty bad, and once you add codec2 modulation
>>> and demodulation to that, it may not come across very well. Just some
>>> food for thought...
>>>
>>> Cheers,
>>> Mark VK5QI
>>>
>>> On Thu, Jul 23, 2015 at 11:11 AM, David Rowe <da...@ro...
>>> <mailto:da...@ro...>> wrote:
>>>
>>>      Hi Onno,
>>>
>>>      Yayy, the mailing list is working again!
>>>
>>>      Good questions.
>>>
>>>      If you are OK with Linux (and when codec2-dev SVN is alive again),
>>> you
>>>      can use the command line freedv_tx/freedv_rx programs to process a
>>> wave
>>>      file:
>>>
>>>          (build codec-dev as per readme)
>>>          $ cd build_linux/src
>>>          $ ./freedv_tx 1600 speech.wav modem_signal.wav
>>>          $ ./freedv_rx 1600 modem_signal.wav speech_decoded.wav
>>>
>>>      speech.wav should be sampled at 8kHz, 16 bit integer (short)
>>>      samples, mono.
>>>
>>>      Or send me a wave file you want processed and I'll run it for you.
>>>
>>>      Cheers,
>>>
>>>      David
>>>
>>>
>>>      On 23/07/15 09:42, Onno Benschop wrote:
>>>      > Hi all,
>>>      >
>>>      > While Andrew Smith (VK6PDQ/VK6AS) is on holiday I have his
>>> SM1000 to
>>>      > play with. Nice piece of kit. (I added 4 feet to it, makes it
>>> stay on
>>>      > the bench -
>>>
>>>> http://www.bunnings.com.au/madico-surface-protection-8mm-vinyl-rnd-clr-23260cf_p3979381)
>>>
>>>      >
>>>      > For an episode* of "What use is an F-call?" - I want to achieve the
>>>      > following:
>>>      >
>>>       >   * Record some audio that has been transcoded using CODEC2, so
>>> that
>>>      >     listeners get a sense of the difference between analogue and
>>> digital
>>>      >     audio.
>>>       >   * Record some digital - non-decoded - audio, so listeners can
>>> hear
>>>      >     what CODEC2 sounds like on-air.
>>>       >   * Record some digital - non-decoded - audio and put it as a
>>>      >     "sub-audible" signal underneath some normal analogue audio, so
>>>      >     listeners get two information streams for the price of one.
>>>      >
>>>      > I am unclear what audio levels come and go from the SM1000 and I
>>> don't
>>>      > want to blow up either ports on the unit or my computer.
>>>      >
>>>      > I have installed FreeDV onto a Debian Wheezy virtual machine.
>>> (Running
>>>      > on a Mac, so no "microphone port", just a "line level port".
>>>      >
>>>      > I also have a Heil headset and an audio mixer.
>>>      >
>>>      > I've been trolling the documentation, the wiki and other
>>> information,
>>>      > but I'm not clear on how to achieve this electrically.
>>>      >
>>>      > Any pointers would be gratefully accepted. (And when I achieve
>>> it, I'll
>>>      > document it, so I won't forget how to do it later ;-)
>>>      >
>>>      > I also came across instructions to compile CODEC2 from source,
>>> but I
>>>      > cannot seem to checkout from the repository. Not sure if that is
>>> because
>>>      > of the SF outage
>>>       >
>>>
>>> [http://sourceforge.net/blog/sourceforge-infrastructure-and-service-restoration-update-for-722/],
>>>
>>>      > or because of a configuration problem at my end.
>>>      >
>>>      > Look forward to hearing from you.
>>>      >
>>>      > * Technically, "What use is an F-call?" no longer exists. As of
>>> a few
>>>      > months ago it's now called "Foundations of Amateur Radio", but the
>>>      > national news is about a year behind, so nationally this won't
>>> be heard
>>>      > for a little while.
>>>      > --
>>>      > Onno Benschop
>>>      >
>>>      > ()/)/)()        ..ASCII for Onno..
>>>      > |>>?            ..EBCDIC for Onno..
>>>      > --- -. -. ---   ..Morse for Onno..
>>>      >
>>>      > If you need to know: "What computer should I
>>> buy?"http://goo.gl/spsb66
>>>      >
>>>      > ITmaze   -   ABN: 56 178 057 063   -  ph: 04 1219 8888   -
>>>       > on...@it... <mailto:on...@it...>
>>>      <mailto:on...@it... <mailto:on...@it...>>
>>>       >
>>>       >
>>>       >
>>>
>>> ------------------------------------------------------------------------------
>>>
>>>       >
>>>       >
>>>       >
>>>       > _______________________________________________
>>>       > Freetel-codec2 mailing list
>>>       > Fre...@li...
>>>      <mailto:Fre...@li...>
>>>       > https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>>>       >
>>>
>>>
>>> ------------------------------------------------------------------------------
>>>
>>>      _______________________________________________
>>>      Freetel-codec2 mailing list
>>>      Fre...@li...
>>>      <mailto:Fre...@li...>
>>>      https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>>>
>>>
>>>
>>>
>>> ------------------------------------------------------------------------------
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> Freetel-codec2 mailing list
>>> Fre...@li...
>>> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>>>
>>
>>
>> ------------------------------------------------------------------------------
>>
>>
>>
>> _______________________________________________
>> Freetel-codec2 mailing list
>> Fre...@li...
>> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> Freetel-codec2 mailing list
> Fre...@li...
> https://lists.sourceforge.net/lists/listinfo/freetel-codec2
>