[Freetel-codec2] Tone and mixed speech with tone showing very low MOS value

[Suresh Thiriveedi]

Hi,

We want to add the codec2 to our product and I just downloaded the latest
code. I built and tried our sample pcm files (typically we use these files
to compare MOS and bit exactness among various codecs) to get the MOS
scores. I'm surprised to see that for tone and mixed speech with tone and
noise sample, the MOS score is very less as below.

For the normal voice samples i got MOS score around 2.7 for 2.4K and 2.8
for 3.2K. But for sine_wav I got 0.346 for 3.2K and -(minus)0.182 for 2.4K.
Where as for the mixed speed with noise and tone, got 1.651 for 2.4K and
1.220K for 3.2K.

Is there any known limitation for the tone transcoding in codec2? For
voice, it is comparable with AMBE but for tone, it is not. Far
difference.Could someone please comment?

Attaching the sample files for the reference.

Thanks
Suresh

[David Rowe]

Hi Suresh,

How did you measure the MOS scores? Good to see we are comparable to
AMBE on clean speech.

There is no algorithm in Codec 2 to handle tones or background noise, so
it tends to fall over. These low it rate codecs are highly optimised
for speech so need special code to handle non-speech signals. For
FreeDV we use the Speex noise reduction in front of Codec 2 which helps
with noise.

It's something I'd like to add in the future, pls contract me directly
if your company would like to sponsor development of these features.

Cheers,

David

Post by Suresh Thiriveedi
Hi,
We want to add the codec2 to our product and I just downloaded the
latest code. I built and tried our sample pcm files (typically we use
these files to compare MOS and bit exactness among various codecs) to
get the MOS scores. I'm surprised to see that for tone and mixed speech
with tone and noise sample, the MOS score is very less as below.
For the normal voice samples i got MOS score around 2.7 for 2.4K and 2.8
for 3.2K. But for sine_wav I got 0.346 for 3.2K and -(minus)0.182 for
2.4K. Where as for the mixed speed with noise and tone, got 1.651 for
2.4K and 1.220K for 3.2K.
Is there any known limitation for the tone transcoding in codec2? For
voice, it is comparable with AMBE but for tone, it is not. Far
difference.Could someone please comment?
Attaching the sample files for the reference.
Thanks
Suresh
------------------------------------------------------------------------------
_______________________________________________
Freetel-codec2 mailing list
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

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

[Steve]

I might be mistaken, but I think tones are a separate algorithm. For
example, in AMBE the tones are detected and encoded into a digital word
(separate from the speech) and then recreated at the other end. That is,
the tones don't pass through.

[Steve]

Well, disregard my last, as I was only thinking of DTMF tones, not the
tones you are describing.

[Bruce Perens]

The difference between Codec2 and those older codecs for which mixed tone
and voice tests were written is that Codec2 is a pure voice codec. That's
one reason that Codec2 has so much lower bandwidth. I've gotten a few tones
through it, but only ones that sound like a voice.

Post by Steve
Well, disregard my last, as I was only thinking of DTMF tones, not the
tones you are describing.
------------------------------------------------------------------------------
_______________________________________________
Freetel-codec2 mailing list
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

[David Rowe]

Couple of issues:

1/ we don't want the codec to break down when the input audio is
corrupted by tones and/or noise, or various other impairments like poor
quality audio, echo, a siren, two voices etc. By "break" I mean render
the wanted speech signal unintelligible or horrible to listen to. This
is DSP problem with no impact on the bit rate.

2/ In some cases (e.g. DTMF, PSTN call progress tones), we would like to
faithfully reproduce the tones at the output of the codec. In this case
faithful reproduction of input speech is not required for the duration
of the tone (we don't talk while we a dialing). This requires a trivial
increase in the bit rate, like 1 bit/frame to say "this a DTMF frame or
a speech frame". You then have every other bit at yr disposal to
describe the tone for that frame - plenty.

3/ A corner case might be audio from an emergency services vehicle,
where we would like to hear the siren in the background (perhaps at
reduced level) as well as the voice of the person speaking.

- David

Post by Bruce Perens
The difference between Codec2 and those older codecs for which mixed
tone and voice tests were written is that Codec2 is a pure voice codec.
That's one reason that Codec2 has so much lower bandwidth. I've gotten a
few tones through it, but only ones that sound like a voice.
Well, disregard my last, as I was only thinking of DTMF tones, not
the tones you are describing.
------------------------------------------------------------------------------
_______________________________________________
Freetel-codec2 mailing list
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
------------------------------------------------------------------------------
_______________________________________________
Freetel-codec2 mailing list
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

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

[Bruce Perens]

We are planning to use Codec2 on AMSAT's new geostationary satellite, which
is being built in cooperation with FEMA, the U.S. federal emergency
services department. It will at some point be nice to be able to operate in
the classical emergency situations of chainsaws and sirens. IMO filtering
out things that aren't voices would be best. But if you want to move them
to the closest bucket and make them look like a voice harmonic, that might
work too.

Post by David Rowe
1/ we don't want the codec to break down when the input audio is
corrupted by tones and/or noise, or various other impairments like poor
quality audio, echo, a siren, two voices etc. By "break" I mean render
the wanted speech signal unintelligible or horrible to listen to. This
is DSP problem with no impact on the bit rate.
2/ In some cases (e.g. DTMF, PSTN call progress tones), we would like to
faithfully reproduce the tones at the output of the codec. In this case
faithful reproduction of input speech is not required for the duration
of the tone (we don't talk while we a dialing). This requires a trivial
increase in the bit rate, like 1 bit/frame to say "this a DTMF frame or
a speech frame". You then have every other bit at yr disposal to
describe the tone for that frame - plenty.
3/ A corner case might be audio from an emergency services vehicle,
where we would like to hear the siren in the background (perhaps at
reduced level) as well as the voice of the person speaking.
- David

Post by Bruce Perens
The difference between Codec2 and those older codecs for which mixed
tone and voice tests were written is that Codec2 is a pure voice codec.
That's one reason that Codec2 has so much lower bandwidth. I've gotten a
few tones through it, but only ones that sound like a voice.
Well, disregard my last, as I was only thinking of DTMF tones, not
the tones you are describing.

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

Post by Bruce Perens
_______________________________________________
Freetel-codec2 mailing list
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

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

Post by Bruce Perens
_______________________________________________
Freetel-codec2 mailing list
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

------------------------------------------------------------------------------
_______________________________________________
Freetel-codec2 mailing list
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

[David Rowe]

I just had a pleasant across town chat with Peter, VK5APR on 2M FreeDV.
We used SSB radios. Unlike HF no fading ...... so FreeDV 1600 worked
really well and was preferable to SSB due to the lack of background
noise. It also operated noise free at 500mW with a SNR of 10dB. SSB
was hard to read at the same power due to the channel noise, and even at
50W the channel noise was still there.

Interesting to contrast with HF.

Anyway, I'll be working up some FreeDV VHF modes over the next few
months. One that will work with standard FM radios (even $40 HTs no
data port rqd), and another that will outperform FM and 1st generation
DV systems by 10dB.

Cheers,

David



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

[Helmut]

Ufb, David! Did you use mode 700B or 1600?

73, Helmut, DC6NY

-----Ursprüngliche Nachricht-----
Von: David Rowe [mailto:***@rowetel.com]
Gesendet: Freitag, 6. November 2015 04:30
An: freetel-***@lists.sourceforge.net
Betreff: [Freetel-codec2] VHF FreeDV

I just had a pleasant across town chat with Peter, VK5APR on 2M FreeDV.
We used SSB radios. Unlike HF no fading ...... so FreeDV 1600 worked
really well and was preferable to SSB due to the lack of background
noise. It also operated noise free at 500mW with a SNR of 10dB. SSB
was hard to read at the same power due to the channel noise, and even at
50W the channel noise was still there.

Interesting to contrast with HF.

Anyway, I'll be working up some FreeDV VHF modes over the next few
months. One that will work with standard FM radios (even $40 HTs no
data port rqd), and another that will outperform FM and 1st generation
DV systems by 10dB.

Cheers,

David



----------------------------------------------------------------------------
--
_______________________________________________
Freetel-codec2 mailing list
Freetel-***@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2


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

[David Rowe]

We used 1600, 700 and SSB.

Post by Helmut
Ufb, David! Did you use mode 700B or 1600?
73, Helmut, DC6NY
-----Ursprüngliche Nachricht-----
Gesendet: Freitag, 6. November 2015 04:30
Betreff: [Freetel-codec2] VHF FreeDV
I just had a pleasant across town chat with Peter, VK5APR on 2M FreeDV.
We used SSB radios. Unlike HF no fading ...... so FreeDV 1600 worked
really well and was preferable to SSB due to the lack of background
noise. It also operated noise free at 500mW with a SNR of 10dB. SSB
was hard to read at the same power due to the channel noise, and even at
50W the channel noise was still there.
Interesting to contrast with HF.
Anyway, I'll be working up some FreeDV VHF modes over the next few
months. One that will work with standard FM radios (even $40 HTs no
data port rqd), and another that will outperform FM and 1st generation
DV systems by 10dB.
Cheers,
David
----------------------------------------------------------------------------
--
_______________________________________________
Freetel-codec2 mailing list
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
------------------------------------------------------------------------------
_______________________________________________
Freetel-codec2 mailing list
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

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