So a few people have been wondering how I do this, and why I do this. So again to start the mp3 file of the podcast is massive in about 90MB. It is recorded as 96KBPS, 44Khz and in stereo(mp3/wav), and 768kb is only about a minute!

Now running the following program:

sox $1.mp3 -t wav -c 1 -r 16000 - | speexenc --agc --denoise -w --vad --dtx --abr 16 - $1-wide-agc-denoise-16000-16.spx
From within OSX (I would imagine any *NIX could do this, or probably run this on win32 too, the key thing is to resample the mp3 at 16000Hz), And using the speexenc program that is included with the source to the codec.

So this produces a file that is 54kb in size with a rate of 7kbps at 16khz (speex/wav). So yes it does sound .tinny. it isn.t perfect but at about 1/14th the size, it.s a great thing for people on constricted bandwidth, or that don.t want to keep GB.s of sound recordings.

The next thing to try was a more .extreme. or narrow version of the above, this time running this command:

sox $1.mp3 -t wav -c 1 -r 8000 - | speexenc -n --vad --dtx --abr 6 - $1-narrow-abr-6-8000-8.spx

Now this generated a 29kb file, with a rate of 4kbps at 8Khz. (speex/wav) Ok it does sound nothing like the original mp3, but you can still understand what they are saying. And I know for a fact it works over a dialup internet connection at 9600kb!

For those who care, here is a reading of Neuromancer by William Gibson, and the talk radio of Vice City.

Death Throes of the Republic VI
It's from hardcore history, 5 HOURS 22 MINUTES long... 16MB compressed!

Or even the Apollo 11 moon landing!

Elon Musk Talks About Colonizing the Galaxy

Based on ffmpeg

ffmpeg -i 1609_-_Elon_Musk-2aB2swgyXqbFA06AxPlFmr.mp3 -af "pan=mono|c0=c1" -ar 8000 "1609_-_Elon_Musk-2aB2swgyXqbFA06AxPlFmr.spx"
Back