I just purchased the r8brain pro. I feel that it is the best samplerate converter, software-wise, I've ever tried out.
My projects are bounced out of Sonar. It only allows a max of 24 bits of the output wordlength files. (Eventhough the audioengine is 32 bit float - stupidity!) I work in 48 kHz.
Are there any advantages when converting my files (48k to 44.1k) to expand them to 32 bits before, let's say, final dithering in wavelab? I was thinking if you, or someone had got any idea if there is any truncation-like things happening at the lower bits when using r8brain's sample rate conversion. I guess it shouldn't if the wordlength is not reduced (?).
Naturally, there is no furter processing to the file after r8brain has converted it - except from the "down to 16 bits"-dither.
It should be beneficial to convert to 44.1kHz/32-bits since this allows to bypass r8brain PRO's internal dithering stage which is not noise-shaped. Other than that dithering inside r8brain PRO should be sufficient for most uses - so there should be no worries about its quality (i.e. no plain truncation is taking place there).
I'm glad r8brain PRO's performance impresses you.
I've happened to come across this piece of information at the PSW forums:
"The SRC algorithm has its own maximum capable wordlength. In a floating point DAW that might be 32 bits, so if you start with a 24 bit 96K file and then SRC it to 44.1 K, you should SRC it to 32 bit float. And then dither it to 16 bits.
This will produce the minimum error signal and minimize cumulative distortion."
It's the audio guru Bob Katz' opinion about samle rate converison using software.
To connect to this threads previous posts...
I totally agree with your opinion Aleksey. And I love r8brain. The best thing for me has been to input something, export it at 32 float after SRC, and later have a possibility to choose a noiseshaped dither, depending on what kind of curve that fits the material best.
I guess that Mr Katz point out that a SRC can somewhat change an envelopes amplitude and raise peaks with small amounts of 0.X dB. Looking at it that way, naturally it sounds realistic to use all the bits for possible, tiny, gainchanges during the SRC.
I guess that r8brains dither (if going to 16 bits) is a "secondary step" in the total process when going from high SR/bitdepth to a lower, right? And everything should be just fine if it is so.
SRC, indeed, can change the envelope of the peaks. In min-phase mode SRC can create 3 dB or higher peaking higher than what there was in the original file.
In r8brain PRO dither is applied after actual conversion took place. It's not the part of the conversion process.
Ah, up to 3 dB! Well, this certainly indicates that it's a wise choise to have the file at 32 float on the SRC's output - even if the input is 24 or even 16, since it's actually expanding in amplitudes. If it's going to be dithered/processed somewhere else, that is.
I feel a deep respect for your creations, and your knowledge mr. Vaneev. Not only do you come up with fresh and oh-so musical ideas - you also pay a lot of time to support your user-base through this forum. In times of silent, non-responsive and echoing support-departments, I feel a lot of confidence in buying your delicate plugs!
It's also suggested to peak-normalize file to -6 dBFS before doing min-phase conversion so that no peaks actually go over 0 dBFS. Even if floating point cannot clip above 0 dBFS, actual audio application where you are going to load this file may clip it.
Thanks for your support!
This topic was last updated 180 days ago, and thus it was archived. Replying is disabled for this topic.