You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[BUG][TGL]Arecord sound have some overshoot noise output when captured via DMIC0,4. #1876
@sinahuang Can you please attach wav files for both 48 kHz and 16 kHz capture. The screenshots prove the issue but are not as useful for considering fix alternatives as wav files.
It's known issue that there's some remaining DC pulse with 16 kHz DMIC. The start waveform with competing microphone DC pulling up the waveform and lower rate updated gain ramp pulling it down creates some small buzz sound. It should be at least made non-audible.
There's a HW shortcoming with time constant of DC blocker filter that becomes rather slow for low sample rates like 16 kHz. It's possible hide this issue with FW driver even more than currently with gain ramp and apply higher cutoff FW high-pass filtering but the muted time is becoming quite long and might disturb use cases (utterance start could be missed in recording). Though in the plot the DC pulse looks quite strong. Maybe there is no high-pass filtering in topology at all. If that's the case then the 16 kHz case could be improved without increasing muted time.
I got them, the 48 kHz version looks quite normal. Though the 4ch volume control issue shows up.
In the 16 kHz version the voice pitch sounds elevated (another potential issue - please check this). Does the 6.225 seconds recording length match your arecord command? Also if you run "time arecord -d 10" does the shown time match the recording duration?
Yep, there's some DC left in both. I'll see what could be done.
@xiuli Is there in CI test for actual playback and recording lengths, e.g. aplay/arecord 10s should take min 10s and max 11s.? This kind of regressions would be easy to avoid with such test.
@sinahuang Thanks for confirm. The issue #1969 has some similarity. I now have a device that is similar to that report. I will try to enhance capture beginning within firmware what can be done, and I think it could help with this issue with different codec IC too. There could be also issues in driver control sequence.
The issue no more happens with a device that uses sof-cml-rt711-rt1308-rt715.tplg.
Kernel Branch: topic/sof-dev
Kernel Commit: 0253ee5
SOF Branch: master
SOF Commit: 52e82f4
Here's the result of command
$ time arecord -Dhw:0,4 -f dat -d 10 rec04.wav
Recording WAVE 'rec04.wav' : Signed 16 bit Little Endian, Rate 48000 Hz, Stereo
real 0m10,481s
user 0m0,013s
sys 0m0,016s
The recording is at correct pitch, recording length, and no more DC pulses in the beginning. The DC pulse is suppressed by the high-pass IIR filter in capture pipeline.
Activity
plbossart commentedon Mar 9, 2020
@sinahuang do you also see this with the regular 48kHz capture?
Wondering if the filters @singalsu worked on are enabled in 16kHz mode, or if they need to be different?
sinahuang commentedon Mar 10, 2020
Issue still can be reproduced with the regular 48kHz capture.
singalsu commentedon Mar 10, 2020
@sinahuang Can you please attach wav files for both 48 kHz and 16 kHz capture. The screenshots prove the issue but are not as useful for considering fix alternatives as wav files.
It's known issue that there's some remaining DC pulse with 16 kHz DMIC. The start waveform with competing microphone DC pulling up the waveform and lower rate updated gain ramp pulling it down creates some small buzz sound. It should be at least made non-audible.
There's a HW shortcoming with time constant of DC blocker filter that becomes rather slow for low sample rates like 16 kHz. It's possible hide this issue with FW driver even more than currently with gain ramp and apply higher cutoff FW high-pass filtering but the muted time is becoming quite long and might disturb use cases (utterance start could be missed in recording). Though in the plot the DC pulse looks quite strong. Maybe there is no high-pass filtering in topology at all. If that's the case then the 16 kHz case could be improved without increasing muted time.
sinahuang commentedon Mar 10, 2020
Attached the 48 kHz and 16 kHz captured wav files. Please notice the beginning of captured files.
captured wav.zip
singalsu commentedon Mar 10, 2020
I got them, the 48 kHz version looks quite normal. Though the 4ch volume control issue shows up.
In the 16 kHz version the voice pitch sounds elevated (another potential issue - please check this). Does the 6.225 seconds recording length match your arecord command? Also if you run "time arecord -d 10" does the shown time match the recording duration?
Yep, there's some DC left in both. I'll see what could be done.
sinahuang commentedon Mar 16, 2020
Checked it seems the recording length doesn't match arecord command .
Run with command "timed arecord -d 10", but actual recording length is 30s,
Wav files:
10s.zip
ClarexZhou commentedon Mar 24, 2020
Blocked by issue thesofproject/sof#2445 which can be fixed by thesofproject/sof#2506, still wait to merge.
sinahuang commentedon Mar 24, 2020
Issue also can be reproduced on TGL-RVP with codec ALC711 and RT1308 in SDW mode.
singalsu commentedon Mar 24, 2020
@sinahuang Thanks for confirming the issue.
@xiuli Is there in CI test for actual playback and recording lengths, e.g. aplay/arecord 10s should take min 10s and max 11s.? This kind of regressions would be easy to avoid with such test.
singalsu commentedon Jun 15, 2020
@ClarexZhou @sinahuang Does this issue still happen?
sinahuang commentedon Jun 16, 2020
Issue still can be reproduced on TGL RVP with codec ALC711 and RT1308 in SDW mode.
singalsu commentedon Jun 16, 2020
@sinahuang Thanks for confirm. The issue #1969 has some similarity. I now have a device that is similar to that report. I will try to enhance capture beginning within firmware what can be done, and I think it could help with this issue with different codec IC too. There could be also issues in driver control sequence.
singalsu commentedon Sep 25, 2020
The issue no more happens with a device that uses sof-cml-rt711-rt1308-rt715.tplg.
Kernel Branch: topic/sof-dev
Kernel Commit: 0253ee5
SOF Branch: master
SOF Commit: 52e82f4
Here's the result of command
The recording is at correct pitch, recording length, and no more DC pulses in the beginning. The DC pulse is suppressed by the high-pass IIR filter in capture pipeline.
singalsu commentedon Sep 25, 2020
I'll close this issue. Please reopen if this is seen with TGL since I can't test with exactly the same configuration.