live audio stream analysis

[unosonic]

sometimes I put a little streambox into a park, forest or similar spaces. It's raspberry pi zero based for best battery life, so no option to run birdnet-pi or birdnet-go. The box sends an mp3 stream to an icecast streaming server, nothing special here. I've thought to do the
birdnet-go analysis server-side, but since only rtsp is or will be supported, I've made a small script w/ ffmpeg and inotify:

#!/bin/bash
# grab http stream STREAM in background, create 15sec long pieces, convert them to 16bit mono wavs in DIR as outx.wav, x: 001, 002,...
ffmpeg -v quiet -i STREAM -f segment -segment_time 15 -acodec pcm_s16le -ac 1 -ar 16000 DIR/out%03d.wav &
# watch for files in folder DIR and analyze them when written, i.e. close_write event occured, delete file when done  
while segment=$(inotifywait --quiet --event close_write --format %w%f DIR/); do
        ./birdnet-go file $segment
        rm -f $segment
done

of course, it would be more convienient if birdnet-go would support also HTTP-based streams a input... ;)