This page is to document the audio recording setup used for OLF 2010 and (possibly) future events.
The plan for 2010
I have selected the Zoom H2 recorder. It has four microphones which permit recording from all directions simultaneously. It is also relatively inexpensive (~$150). It operates on AA batteries or an AC adapter and uses standard SD or SDHC cards.
I plan to set one up at the front of each room. The front side will face the audience, with its narrower but deeper "field of view" hopefully able to pick up questioners who don't use one of the hand-held microphones to ask their questions. The back side will face the speaker and will have no problem picking them up, especially with the amplification. Although possible, I will not be connecting these directly to the sound board. This will avoid having to fiddle with sound levels, and also give us a shot at picking up the aforementioned microphoneless questioners.
My testing has found that the H2 will run for about six hours on a set of two alkaline AA batteries. This will relieve me from having to check up on each room constantly, and I can just replace all the batteries at once during lunchtime. Also, since it uses a FAT filesystem, file sizes are limited to 2GB. This holds three hours of audio. The recorder automatically starts a new file when the limit is reached. Testing showed that with the current firmware (version 1.81), at most three to five seconds of audio is lost in the transition. A 16GB SDHC card will hold about 12 hours of audio (in four-channel mode, two files are recorded simultaneously - one for the front and one for the rear).
As for settings, as stated I am using four-channel mode with all microphones active. The audio is saved as two 16-bit WAV files at a sample rate of 48 kHz (I have read elsewhere that this is the native clock rate of the H2's electronics, and that 44.1 kHz can introduce artifacts). The Mic Gain switch on the side is set at M and the software Rec Level is left at 100. This should prevent any excessive levels that will cause clipping and distortion while still providing good sensitivity (while making test recordings in my apartment with the window open, during quiet periods the microphones recorded crickets chirping outside).
Tomorrow I will order more recorders for a total of six, so we can cover each room. Along with each one I will get a 16GB SDHC card, plus two more extra as spares and for the Friday early penguin and medical tracks. I'll also order an ordinary camera tripod to mount each recorder on and a pile o' AA batteries to keep them fed.
For distributing recordings, I plan to go with archive.org again like last year. I'll produce and upload high-quality mono FLAC format files along with much smaller Speex-encoded files. Cutting the tracks to length and post-processing to normalize levels, remove noise, do dynamic range compression, etc. is done in Audacity. Creating FLAC and Speex files is handled by the olfencode.sh script, available from either SVN or perhaps as a released file. Conversion to Ogg and MP3 from the FLAC sources is handled by archive.org.
How it went
Pretty smoothly, actually. Keeping up with battery and SD card changes was not a difficult challenge. I miscalculated slightly, not realizing that we would record Ubucon on Friday, but we had enough SD cards to cover everything.
Sound quality was mixed. The smaller rooms (Ubucon, Early Penguins, Med Track, OSSS) worked out well - it was even possible to pick up some questioners who didn't use microphones. (They had to be amplified in post-processing to be audible, though.)
Recordings in the large ballrooms did not come out as well, as the signal level from the speakers often struggled to overcome the background noise. Possible solutions are:
- move H2s closer to the speaker
- orient H2s so the microphones point upward toward the PA loudspeakers
In fact, all recordings could have been louder - overdriving the mics was nowhere near a problem, so we should be aggressive in positioning them.
Post-processing and uploading has turned out to be quite time-consuming. This requires more planning next year. We had one audio processing party that got through just over half of the recordings, but lack of time on my part has prevented organizing a second one. Trying to do it in my spare time is causing things to drag out. The labor-intensive part is the Audacity work - the olfencode.sh script basically operates without any intervention needed. It also takes quite a bit of time to upload each FLAC file (~100 MB each).
-Vkochend 03:40, 26 October 2010 (UTC)
How it went for 2011
One major change was made. In the three large ballrooms, we were not able to capture speakers well with the built-in microphones of the H2s, no matter how we oriented them. So instead we plugged the sound boards into the line-in jacks of the H2s. This appeared to give us good levels, but I haven't yet listened to the recordings to be sure this worked well. When doing this, be sure to check the level meter on the H2 ahead of time. In one case, the soundboard was putting out too high a signal and we had to ask the AV company to provide a cable from an output on the board that had a volume control.
The smaller rooms seemed to work out fine as before.
More notes to come as we process the recordings.
How it went for 2013
A few updates. We moved to having a direct line-in connection from the soundboard for all the Saturday recordings. This worked well, but be sure to verify that the signal levels aren't too high. Also, one of the Career Track talks was a panel discussion, but we only had one microphone (it was used for audience questions). Surprisingly, it picked up the panelists from across the room, albeit at very low levels. Major editing was needed to boost them, and the results were still on the low end of "listenable." For future panel discussions, make sure that either all participants have microphones, or stick the H2 on a tripod and use its built-in microphones to catch all discussion in the room.
The AV company did not provide soundboard connections on Friday (due to a miscommunication on our part). Nevertheless, we were able to capture decent audio with the H2s mounted on tripods using their built-in microphones. They were pointed upward to capture sound from the ceiling-mounted loudspeakers. This worked fine in the C-pod rooms; the grand ballrooms have much higher ceilings and it seems like this approach wouldn't work there.
As before, editing and uploading was time-consuming, but we were able to complete this within five weeks. Plan for a total time budget of one hour per talk to cover editing, encoding, and uploading.
Vkochend 00:55, 21 October 2013 (UTC)