04.13.11
TechBytes Encoding
Summary: File sizes to shrink considerably in future shows
TIM and I had a quick chat yesterday after it turned out that we had oversampled 38 shows/episodes, basically saving at a sample rate far greater than what we had recorded. This becomes especially problematic because audio files are serves by a cache server that drops the connection after 10 minutes, which means that timeouts will affect downloaders on a low-bandwidth connection. Some listeners reported this to us last year, but it wasn’t until yesterday that we thought about shrinking file sizes. The idea came from a listener. On an important and technical note, this listener asked whether feasible to “make the files much smaller (monophonic with a slower sample rate and lower resolution).” I then spent over an hour experimenting with Audacity and ffmpeg
. I ran a series of tests where parameters in the encoders are changed and I then listened to the results. The VoIP recording we do is at 16KHz, compared to the far greater frequency which we used to encode this in (huge waste of space for no gain in quality). Eventually I found that even without using the command line, e.g. ffmpeg -ab 16k -ar 11025
, I could still rely on Audacity to handle both .mp3
and .ogg
extensions to produce files about a quarter of the size they used to be. The only drawback of that is that when we have music tracks (of higher sampling/resolution) they obviously degrade in quality. Since the vast majority of the recorded tracks will always be talk, that ought to be worth the compromise and smaller files would also make storage and bandwidth less of an issue (caching resolves only the latter, to a degree).
Would anybody be interested in video versions of the show? SIP Communicator and Ekiga are compatible with my webcam and gtk-recordmydesktop
can be used to show Web sites and other things while the show is running. █
satipera said,
April 13, 2011 at 4:32 am
Yes I would be interested, body language, facial expressions etc add much more to communication.
Dr. Roy Schestowitz Reply:
April 13th, 2011 at 8:47 am
I was thinking about visual illustrations more than about faces.
Needs Sunlight said,
April 13, 2011 at 4:33 am
The audio-only is enough. Since it is the discussion that is the content, adding video would not be worth the extra efforts and bandwidth.
twitter Reply:
April 13th, 2011 at 8:03 am
If that were true, the IRC logs would be better. If audio is worth while, so is video.
Dr. Roy Schestowitz Reply:
April 13th, 2011 at 8:46 am
This can be used for showing Web pages and stuff (maybe faces) while we go along.
twitter Reply:
April 13th, 2011 at 6:50 pm
Sounds great. I was imagining a recording of a video meeting using gtk-recordmydesktop or similar. Everyone should be able to see everyone else’s face or avatar for the meeting and you can run a browser and IRC chat too. If nothing else, that would be a cool meeting. The resulting ogv does not have to be too outsized, and you can host it as a torrent.
Are the free software sip clients up to this yet? Skype can do the multi video chat thing I think, but it is a pay service and I don’t think Skype works with Lenny or on anything but x86.
Dr. Roy Schestowitz Reply:
April 13th, 2011 at 8:46 am
I meant separately from the audio.
kozmcrae said,
April 13, 2011 at 8:50 am
I don’t think video is worth the bandwidth unless there are some very interesting visuals to see or your guests each episode include people like; Mila Kunis; Natalie Portman; Maggie Gyllenhaal… You get the idea. Second thought: Maybe once and a while a video episode (with an audio only option) would be a good fit for the material.
No, seriously. It’s good to hear about the bandwidth. I would also like to add my two cents about the sound. First, the volumes between the three of you are uneven. I used the compressor in Audacious and was able to even it out somewhat. But there’s one aspect of the sound I can’t seem to control. It’s the relative frequencies. Tim comes through the best, smooth and well rounded. Then Gordon’s, but Roy, you must burdened with a poor quality microphone because your voice sounds like it’s coming through on one band in the midrange of an equalizer. There’s no way I can fatten it up to sound more like Tim’s voice. I spent the better part of my life playing music and working with sound, maybe that’s why I’m sensitive to it. Unfortunately I no nothing about working with sound over the Internet so I have no advice, just some critiques.
Dr. Roy Schestowitz Reply:
April 13th, 2011 at 9:20 am
Thanks for the advice. When it comes to sound volume anomalies, these change between episodes depending on distances. We can try to adjust that in advance. I haven’t checked to see if there’s a filter that can detect particular voices and calibrate volume depending on the person (or just equalise all).