Some technical notes on our lecture recordings

We now have a digital sound recorder built in to our sound system, which records mp3 files straight to a flash drive.

Formerly, we used a Tascam DR-01 digital recorder and a Shure SM-58 microphone to record lectures at the Vedanta center. We used to plug the recorder right into the sound system, but our wireless mikes were dropping out from time to time, and the amplifier was adding a low but audible hum. This way we have an extra microphone at the lectern, but we get a clean recording. When recording classes in the living room, we use the built in stereo microphone on the Tascam recorder, and hide the recorder up on the mantelpiece. We are recording in mp3 format at 256 kbps, rather than wav format, to reduce file size. Though mp3 is a lossy format, we thought that these lectures were not so important that we need to record in wav or another non-lossy format, especially since we are recording at a high rate.

Audio processing: we use the free open source program Audacity to prepare the lectures for online podcasting. It’s a great program, though it takes some time to learn how to use it. We take the following steps.

  • First, amplify (the microphone is not boosted, so recording levels are very low).
  • Remove background hiss via noise reduction.
  • Apply compression (to even out the sound volume).
  • Apply a hard limiter to catch any high points that the compressor didn’t quite level out.
  • Amplify again to -1.5 dB.
  • Fade in at beginning, fade out at end.
  • Remove any coughs etc that we can catch (but we don’t listen to the whole thing—often one can see such sounds by their very loud volume).
  • Remove the announcements and other unimportant stuff.
  • Export to 256 kbps for our archive, and again at 64 kbps for upload to the website.

Advanced functions include the ability to remove hums or tones of specific frequencies. This is great for removing electrical hum, and also for knocking out feedback. We have been able to save some poor recordings spoiled by feedback, using the Nyquist notch function. First, one analyzes the feedback (plot spectrum) to determine the exact frequency, then one runs the notch over the whole recording, and the feedback is gone.

The Nyquist notch format for single, mono tracks: (notch2 s * Q), where * is the frequency to be notched, and Q is a number which determines the “depth” of the notch. We have found 50 to work well; the lower the number, the deeper the notch. For example, an electrical hum in the USA will have a primary frequency of 60 Hz, so our notch (via the Nyquist prompt) would be (notch2 s 60 50).

Typically a hum will have several frequencies; electrical hums might have frequencies at 60, 180, and 300 Hz. To run the notches simultaneously, the following code can be used:
(notch2
(notch2
(notch2 s 60 50)
180 50)
300 50)

The frequencies to be notched can be determined from the “plot spectrum” function. One selects a very short (.25 sec or so) portion of the track in which the tone to be removed is dominant, analyzes it by “plot spectrum”, viewing the “log frequency”.

For stereo tracks, the notch format is:
(vector (notch2 (aref s 0) * Q) (notch2 (aref s 1) * Q) )