Some questions on transcoding

[Lex]

Hello Simon,

I hope you could clarify a few things for me. My situation is that I am using REW-created IR convolution files together with transcoding:

- After convulution the volume is about 8 dB lower than before. In my version of ffmpeg the gtype=gn function still works, but I found that in some cases the replay gain of the stream is actually increased.
I have tested a file that originally that had peak value of 1.00000, which, after convolution had a higher replaygain. This makes me wonder whether the gtype=gn command could result in clipping. Would it not be better to use another command to increase to volume with a fixed conventional gain of, say, 2?

- Whatever setting I choose for DSF to PCM conversion, the sound is always affected. Since I want the room correction also for DSF, this is an issue (as my Linn DS3 natively handles DSD). I asked Gemini, who came with a quite interesting recommendation which I would want to try. Unfortunately, I don't know how to enter this in mminimstreamer.
The idea is that DSD produces lots of HF noise, and we would want to include a low pass filter at 28 kHz before transcoding to PCM. The suggested PCM sample frequency is 88 kHz. A restoration of the volume with 6 dB is also suggested (although I would prefer a factor of 2 over 6 dB since that would avoid rounding errors).

- Apparently you can set the precision of the afir filter to "double". Would that yield more accurate results?

Thanks in advance for your response!

[simoncn]

MinimStreamer transcoding works by running FFmpeg with filter options provided by the user in the stream.options property. These options are very complex and I don't have the advanced knowledge of FFmpeg and transcoding that would be needed to answer your questions.

It is possible that someone else with more knowledge of FFmpeg might respond to your post here. Your other option is to direct these questions to the FFmpeg user community or developer community. To enable this, I suggest you try running FFmpeg as a standalone command using a small number of sample audio files that illustrate the problems you are having. If you are able with help from the FFmpeg community to find the correct FFmpeg options to make this standalone FFmpeg transcoding work correctly, I will do my best to help you to apply these options to MinimStreamer.

[Lex]

Hello Simon,

I tried your suggestion and spent way too much time trying different settings. But, whatever settings I use, it does not seem possible to retain the full audio quality converting DSD to PCM.
I tried all target sample frequencies (88, 176 and 352 kHz), different lowpass filters (4th order LR at 40 kHz would be my preference, made sure I use sox at 32 bits precision, dithering etc.

It occurred to me that perhaps my DAC is better at playing DSD than it is at playing PCM, but the CD versions also sound better than the converted SACDs, in some ways. Sure, the CD is not as refined and the bass is all over the place (relatively speaking), but some of the space and transient edge of percussion etc. seems to get lost in the conversion process.

I believe I must accept this, and make a choice if I prefer the room correction, or the unrestrained sound quality of the DSD file. At least: I won't look into this any further.

[simoncn]

This was my experience as well. I have purchased the same music in DSD and PCM and untranscoded DSD was clearly superior to PCM or transcoded DSD. This was true even when comparing untranscoded DSD with untranscoded hi-res PCM (24/192).

[Lex]

Now that you got me playing with ffmpeg, which is VERY interesting, I noticed the following. I put a 24 bit WAV file through the afir filter and the output was in 16 bit. I needed to set the output type to get 24 bit. I only found the issue by chance after unexplained loss of sound quality, a lot of fiddling when I noticed the file size being too small.

Since you said somewhere that first there is the convolution and then the signal is put through the resampler (I use flac:wav24 semicolon), I now worry that in the intermediate step the resolution may be reduced to 16 bit. If you did not manage that, I think it would be the case.

I don't think this actuallly happens, since I (think) would have heard it, but it would be reassuring if you could confirm that this does not happen.

btw. I was looking for the best way to manage the signal level after convolution and it seems the best way is to leave irnorm and irgain and, especially, dry at their default values and use wet=2, which is bit-for-bit identical with ,volume=2 (I checked).

[simoncn]

You can add -v 100 at the start of stream.options to see exactly what intermediate streams are being created. This option adds detailed ffmpeg log output to the MinimServer log. I know that I came across this issue and made sure that the intermediate stream wasn't using a lower resolution than the input to it.

[Lex]

Hello Simon,

I cannot seem to stop comparing the different settings in ffmpeg! What I noticed is, when I use double precision for the afir command, the sound seems a little more clear/ transparent. I cannot really find out how all the ffmpeg commands work exactly. Maybe the only way to do that is to study the code. I wish that more info was available. I say this because it would be nice to check if there is indeed a technical limitation to the default settings.

That said, I don't think it is currently possible to use double precision in Minimstreamer. I wonder if it would be possible for you to add that option. The command would probably come to look like this:

convOut=-i /volume1/MinimServer/opt/bin/filter*.wav -lavfi afir=wet=2:precision=double

[simoncn]

Please post a zip file containing the filters you are using.

[Lex]

(11-10-2026 15:45)simoncn Wrote:  Please post a zip file containing the filters you are using.

Not sure what you mean. The command I used in powershell is this:

ffmpeg -i name.wav -i filter.wav -filter_complex "[0:a][1:a]afir=precision=double, volume=2" -c:a pcm_s24le output.wav

Since volume does not seem to work in minimsteamer, but wet does, I am looking to create the following command in minimstreamer:

ffmpeg -i input.wav -i filter.wav -filter_complex "[0:a][1:a]afir=wet=2:precision=double" -c:a pcm_s24le output.wav

Not sure if this what you asked though.

[simoncn]

In your previous post you were using these filters:

/volume1/MinimServer/opt/bin/filter*.wav

Can you post a zip file containing all these filters for different sample frequencies? This will enable me to test with your exact configuration.