I record primarily birds at 48kHz, medium sensitivity. I have noticed that with Audiomoth 1.2, I can throw away 5–6 least significant bits without any noticeable effect on spectrograms or in sound quality. In practice, when I compress to FLAC, I do something like
ffmpeg -i $file -filter:a "volume=-36dB" -sample_fmt s16 -acodec flac `basename $file .WAV`r.flac
and then archive the file (or run Birdnet LITE, or look at spectrograms with renormalisation after loading the file). The "-36dB" is there to discard six bits. sox can do the same, but it discards headers, therefore ffmpeg. In compression, this saves a lot of space (over 50%), because the last bits are essentially random.
Has anyone compared the SNR of 1.1 vs. 1.2 on audible frequencies? That is, what number am I supposed to use in place of -36?