Hi all, I guess that this is related to the post "HifiBerry Digi - ALSA supported formats"... anyway, I'm trying to record an audio stream from my adc (96 KHz/24bits) through the S/PDIF input, but I'm unable to do it using a 24 bits format:
arecord -d 0 -r 96000 -c 2 -f U24_LE test.wav
gives back:
arecord: set_params:1229: sample format not availableavailable formats:S16_LES32_LE
Is there a known way to record using 24bits format?
Thank you
The driver does not support 24bit sample format. You need to use 32bit sample format (S32_LE).
Best regards,
Daniel
Thank you for your reply. I then tried the 32bit sample format (S32_LE) and recorded few test tones plus some tracks.
Despite setting it at 32bit, when I open the saved stream in Adobe Audition CC and run some analyzer, the measured bit depth reported is 20.
The same analog test tones and tracks converted through the same ADC and saved to file with a different device (hence not the Digi+ i/o), report 24 bit as measured bit depth.
Do you know why the DIGI+/arecord saves a 20bit format? Is this some sort of artifact or am I missing something?
Thank you and kind regards,
Antonio
The Digi+ does not save anything. If just sends the data to the application.
I don't really understand what exactly you're doing and how you "measure" bit depth. You're talking about anylog test tones, but the Digi+ is purely digital. Are you comparing some analog signal to a digital signal?
You can measure the dynamic range, but even 20bit will give you around 120db dynamic range which is more than most of the audio playback equipment can deliver (I'm not even starting to discuss real playback with real speakers in real rooms which will reduce dynamic range even further).
Best regards,
Daniel
Hi Daniel,
I know that the Digi+ does not save anything, that's why I have coupled it with arecord in my previous post ;)
What I have is an analog signal that I convert to a digital one with an external ADC (96 kHz/24bit). I then use the Digi+ to receive such digital signal and save it to a wave file via arecord.
I don't want to enter into the discussion of playback quality, speaker quality either, I'd like just to understand the full extent of capabilities of the signal acquisition chain that I'm using to record my signal. I have also to confess that I'm a beginner in the domain of digital audio, so I'm climbing my learning curve. Please excuse me if I'm not clear or I ask trivial questions.
Coming back to my curiosity, what I'm doing is also comparing the digital stream acquired with the Digi+ and saved with arecord with a similar file acquired and saved with different hardware. I'm naively expecting the two to be the same - as the analog to digital conversion is performed by the same unit. However when I run the amplitude analyzer in Adobe Audition CC (is a function that reports, among various measurements on the saved wave file like perceived loudness, peak amplitude and the like, also the measured bit depth), the measured bit depth for the acquisition chain involving digi+ and arecord is always at 20 bit and for the other unit is 24 bit.
Most likely this cuts down in understanding how the bit depth is measured by the software and that's what I'm currently doing. In the meantime, if you have any idea or suggestion, please let me know :)
Thanks,
Antonio
Hi, after some time I'd like to come back on this topic with some additional tests that I made.
The chain that I am testing is
I then open the two wave files created and use a plugin (Bitter from Stillwell) that shows how many of the available bits are being used by the audio signal.The file created by the chain using arecord with hifiberry shows that only 20 bits are used whereas for the other file, Bitter shows all 24 bits in use (screenshots below)
(could not fit two images in the same post)..
Anyway, I was naively expecting to see the same number of bits in use for the two chains. What could cause the chain with the hifiberry to have only 20 bits in use? Any ALSA configuration?
Thank you for your support,
Antonio
Unfortunately, I can't comment or speculate on this.
Best regards,
Daniel
Hello,
further investigations on this issue led me check the WM8804 datasheet. It seems that, by default, there indeed is a truncation of the data word to 20 bits. To mask this truncation the WL_MASK bit in the register R29 should be at 1 instead of at 0 (default).
Is it possible to change this bit from the hifiberry_digi driver or is there a better way to do it?
Kind regards,
Antonio
Antonio Petrella got in touch with me today.
After looking at the register values and measuring the signal transmitted, it is indeed preserving the 24 bits of resolution.
However I was unable to record at 174000 or 192000 Hz sample rate from TOSLINK
Thank you Francois for the useful discussion and tests you have made.
I'm not sure how you compared the signal transmitted with the source, however what I was talking about was the number of effective bits used. Namely the file is always at 24 bits, (or 32, if you use S32_LE) but then with a deeper analysis, the number of used bits was found to be at 20.
Anyway, I made some progress on this issue, and I post the outcome here, if it can be of anyone interest: I have modified the driver and checked the value of WL_MASK. It was at 0. I changed it to 1, recompiled the module and recorded. Then I have analyzed the files with the same tools as I did before. This time the effective bit depth of the signal was at 24 bits.
Kind regards,
Antonio