Hello Ed & list, Thank you for those experiments, very useful data points at this stage in the Codec's development. When I developed the Asterisk support I also learnt a lot by talking to my two sons (different ages and hence pitch ranges) using the codec. Conversational tests show up a lot that you miss when listening to the same utterances all day.
I agree on the buziness artefact - it's the phase synthesis. The voiced speech tend to be synthesised a bit too impulsively (energy concentrated near one position in time). The effect is worse when listened to through a phone handset or headphones, better when through a small laptop speaker (similar to a 2-way radio speaker). Yes, the sharp impulses can be tiring on the ear after a while. I haven't worked out how to fix this yet, but have some ideas. In contrast AMBE codecs tend to have a reverberant or underwater effect to them, especially for males, once again due to the phase synthesis algorithm they use. They sound much better for females, the shorter pitch period makes phase less of an issue. I think the Inmarsat services use a 4 kbit/s AMBE codec. Re the problems you had with high pitched voices, it could be the pitch estimation algorithm breaking, or simply hitting it's limits (I hard coded it arbitrarily to 50 - 400Hz). If I can get some samples of it failing I can track it down. I think Asterisk has a monitor function, I would need the input speech. Curiously my 6 year old didn't have the same problem - in fact he sounded clearer to me than on a GSM phone. The problem with call progress tones is a new one on me, hadn't thought to check that. Thanks. Sounds like we are in the ball park with Iridium at 2400 bit/s. We are behind other codecs at twice the bit rate, but that is to be expected. So not bad for an open source codec that is still under development. Sure Ed - I'll contact you off list to have a call over the codecs...... Thanks, David On Thu, 2012-05-31 at 23:19 +0100, Ed W wrote: > On 31/05/2012 19:25, Ed W wrote: > > Hi Folks > > > > Now that we kindly have an Asterisk codec provided, it's great for > > testing out voice quality. > ... > > I setup prefixes so that I can dial: > > 40xxx dials the number using Codec2 > > 41xxx dials the number using LPC10 > > 42xxx dials the number using G732.1 5kbit > > > > This is great for comparing quality of codecs quickly. > > > I have tested using > - the asterisk echo test > - calling between two snom phones on g711 > - calling between a DECT handset and a nokia wifi phone (g711 & ilbc) > - compared with Inmarsat/Iridium satellite (4/3 kbit ish each) > - Codec2 at 2,500 baud > > My observation are: > - Codec2 is distinctly superior to LPC10. Both are a bit buzzy, but > LPC10 is distinctly more so. Voicing is somewhat similar with each though. > - Codec2 is substantially lower quality than 5kbit G723.1. Actually > quite amazing how good G723.1 is!! (6kbit G723.1 is MOS 3.9 !) > - Codec2 gives me a bit of earache after just a few mins. Not measured > it, but wondering if there is some harmonic distortion generated? > - Unsurprisingly multiple low quality codecs together are additively > poor. Dect + Codec2 has a noticable extra buzziness > - Quite a wide range of voice pitches are handled very nicely. I can > listen to my kids, wife and myself (British english speakers) without > any problems. Upper end of my kids and wife's voice goes distinctly > harsh and buzzy (guys talk in a falsetto to repro). > - Extremely understandable speech with that 8khz kinda feeling, ie not > nasally > - Codec2 deals quite badly with music on hold and ring tones. OK, even > GSM deals badly with music on hold, but the ring tones and other call > progress tones on the PSTN turn into farts and pops (amusing) > > - Compared with Iridium/Inmarsat, those are both less buzzy and somewhat > easier to listen to > - Iridium I believe is somewhere between 2.4 and 3.4 Kbits (not clear > which). It is much more nasally sounding, but it's also smoother > sounding (less buzz). I believe they use a narrower freq filter range > (hence the nasally sound), but the vocoder is less buzzy, on the flip > side Iridium is less easy to hear the non plosive sounds and is often a > little "honky" compared with codec2. Overall codec2 is easier to > understand, iridium is smoother on the ear. > - Inmarsat is far superior to either and usually sounds extremely clear > when the signal is good, normally close to G723.1. However, it also > degrades very well with packet loss and sounds something like Iridium > under stress. I believe they use a 4+Kbit voice codec, but not sure > > Overall seems very impressive. I guess I wish for a touch more voice > quality at the 2,400 level. What I mean by that is that it tends to > oscillate and "buzz" at pitches that are easily found in female and even > high voices - the actual voicing is great, satisfactory identification > of speaker from their voice, clear consonant sounds and non nasally > sounding - excellent. Kids voices are a tough challenge and codec2 does > an admirable job, not much worse than GSM, but I think probably the same > as above - higher pitched voices might be slightly falling outside of > the oscillator prediction area and tend to buzz a little too much? > (could the predictor follow the voice range?) > > I haven't tested in the presence of significant background noise. Which > is to say I stood next to a noisy road and the tv, but it seems the > handsets I own or the input filters are killing the background noise > enough that it wasn't audible, nor a problem at the levels I tried. I'm > presuming this would be a tough challenge on any codec, but it seems > "normal" noise levels are no issue at all. > > > David, do you want to take a voice call over the PSTN using Codec2 just > for kicks? I'm on GMT, propose a time and tel number offlist if so. I > can also call on Iridium/Inmarsat > > Thanks for your wonderful work! > > Ed W > > > ------------------------------------------------------------------------------ > Live Security Virtual Conference > Exclusive live event will cover all the ways today's security and > threat landscape has changed and how IT managers can respond. Discussions > will include endpoint security, mobile security and the latest in malware > threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ > _______________________________________________ > Freetel-codec2 mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/freetel-codec2 ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Freetel-codec2 mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/freetel-codec2
