First test of Codec 2

January 6, 2011 | Amateur Radio | By: Mark VandeWettering

I’ve previously mentioned David Rowe’s excellent work on a patent free codec for amateur radio, Codec2, but today is the first time I actually downloaded the code and gave it a shot. You can find the code and the quickstart here:

Codec 2 « Rowetel

My own voice has some fairly low frequencies in it, so I was intrigued to find out how my own voice would do. David’s page mentions that Kristoff, ON1ARF, has a voice where the existing AMBE voice encoder does a much better job, largely because the low frequency encoding of AMBE is much more effective, so I took a short message that I used for some companding experiments I did a while ago, and gave it a try.

The original, as recorded via my iPhone, and downsampled to 8khz using sox.
The same clip after encoding/decoding with codec2.

As you can probably tell, the fidelity isn’t all that stellar with my own voice either. I have nothing to add other than that base observation, but perhaps this will serve as incentive for me to look into this stuff more.

Comments

Comment from Josh (KD8HRX) Smith
Time 1/6/2011 at 12:56 pm

Mark,
Very cool stuff – codec2 is perhaps the most important amateur related projects going on right now. It’ll really be nice if it can replace AMBE in a DSTAR like network.

73 DE KD8RHX

Comment from David Rowe
Time 1/9/2011 at 5:06 pm

Hi Mark,

Thanks for that test, yes Codec 2 sounds a bit rough with your voice. I can here the pitch estimator making errors. Also the level of your source speech is fairly low compared to some other samples.

All excellent tests. This is what I love about open source DSP development – people stress code in unexpected ways which leads to improvements. This approach also worked very well with Oslec – the open source echo canceller I developed for Asterisk a few years ago.

I’ll use your sample for some more development in a few weeks. By looking at the internal states of Codec 2 when processing your sample I will hopefully be able to improve it.

I must also come up with a simulation for SSB and NBFM to use for comparisons. That is the speech quality target for Codec 2, e.g. it’s not meant to aim a MP3 quality. The upper limit would be cell phone quality.

Cheers,

David