And I added some tests with the Fairphone FP3 I have, using all working profiles.
Test setup:
I used the same audio file, playing back through a Bluetooth HiFi speaker aproximately 2m away from the fairphone and on a different piece of furniture. The test sample was played back at identical volume each time:
Original Test Sample
I was able to make recordings in the following modes:
DEFAULT
and
UNPROCESSED
As you can see, the sources DEFAULT, MIC and CAMCORDER all employing the exact same audio profile. As audible in the test, it employs an auto-gain that fails horribly with music, ruining not just music recordings themselves but also voice recordings with music or other loud noise in the background.
VOICE_RECOGNITION applies additional filters, as can be seen in the spectrogram in reduced high frequencies. However it still employs an auto-gain confused by music and noise, unusable for music sources. The filter also ruins recording of voice with noise in the background.
VOICE_COMUNICATION is the filter normally used during voice calls. It assumes the speaker is directly next to the microphone, so it uses much lower gain to begin with. That doesn’t stop it from being confused by background noise though and lower the gain further. This is probably why people have complained about horrible voice quality. The filter is OK with a single voice source, but if there’s music or party noises in the background, the filter goes GAGA, and whoever is on the other side of the call wouldn’t understand a word.
This is probably the most critical flaw and must be fixed, as it affects using the Fairphone for making Phonecalls!!!
UNPROCESSED has horribly low initial gain, but at least nothing beyond that is applied. The signal is weak, but of high quality.
The gain filter misbehaviour is so pronounced it is even visible in the spectrogram plots.
Look at second 60 to 65 and 155 to 160 - this is always when loud music is encountered, after around 5 seconds, the auto gain goes BERSERK and cuts down gain across a large part of the spectrum, not only ruining the music but also making the speaker completely incomprehensible.
The other profiles could not record in this test. VOICE_PERFORMANCE is probably not yet supported by this Android version, while REMOTE_SUBMIX only makes sense with a remote speaker such as a car connected and VOICE_UPLINK, VOICE_DOWNLINK and VOICE_CALL would only work during a voice call.