It is hard to find in a search engine because “64 bits” is also a very popular keyword about the architecture.
My question concerns the internal audio progressing of Ardour (and NOT the input, output, or file types)
e.g. if I apply a -60dB on an event, and then a +60dB on the channel of this event (let’s say I use the gain), what was the bit-depth of the signal during this processing?
If the program is running with an internal bit-depth of 16, 24 or 32 bits, I will loose a lot of quality in my stupid example.
If the program is running with an internal bit-depth of 16, 24 or 32 bits, I will loose a lot of quality in my stupid example.
No you won’t. In your example it is doubtful you would lose ANY quality, given a couple of exceptions…
That if you are in 16 Bit your signal is in the top 30dB or so of the FS range. (approx 96dB range in a 16 bit signal)
In 24 Bit this increases to the top 80dB or so. (approx 144dB range in a 24 Bit signal). You are more likely to lose resolution from AD/DA process (Which is only about 20bit in most convertors) than from a simple 60 dB boost/cut in 24 Bit audio processing where you remain in the digital domain.
In 32 Bit range this is at a point I don’t feel like doing the math to figure out the range, suffice to say you won’t have an issue.
Jack and Ardour both operate at 32 Bit, as has been mentioned above. Where this really comes into play as a limitation is dealing with track counts more than processing, as in terms of processing this gives a range far greater than that of the human ear. You can run into limitations of the bit depth in processing when dealing with mixing many tracks together, somewhere over 100 IIRC but again I don’t really feel like dealing with the math right now.
So unless you are trying to use Ardour or Jack for something that is unrelated to the limitations of the human ear, or are mixing a LOT of tracks together (And 100 tracks is a fair amount even for me, but does happen fairly common in audio post for video), it is a non-issue. And if the former of those is the cause I would venture there might be more appropriate tools out there.
you may feel there's a very digital sound in the end (cold sound) without knowing where it really comes from.
In a ‘clean’ digital system you won’t get all the ‘character’ associated with some analogue processing - whether that’s a good / bad thing depends upon what you are trying to do - a lot of my plugins are based on analogue designs, and getting some of that ‘analogue’ character is an important part of the process - that’s why we add gentle (tape style) saturation to e.g. the channel compressor plugins, when they are driven at high signal levels, and why the linuxDSP Pultec EQ includes transformer-coupled tube emulation (something you don’t get in many similar emulations for other operating systems, including some requiring extra hardware).
Ironically, in the PEQ-2A we actually use some 64Bit internal (floating point) processing within the plugin, in order to emulate that analogue ‘character’
You may not loose some audible quality but when mixing very sensitive signals (especially with dynamics processing) you may feel there’s a very digital sound in the end (cold sound) without knowing where it really comes from.
That’s usually why I don’t do anything else than volume (automation) in digital and I let the rest, including the summing for external analog devices.
Digital dynamics processing can create artifacts close to the sampling frequency which alias back in to the audible band. Having more bits wouldn’t help with that problem, but the better dynamics processors address it by oversampling so the dynamics processing is done at a higher sampling rate.
It’s 32 bit floating point, which means applying a -60dB cut and then a +60dB boost will never truncate the bit depth; it will just add rounding noise at the 24 bit (way below audible) level.
32 bit float = 24 bit mantissa and 8 bit exponent.
see http://ardour.org/key_features under “Mixing”.
32 bit float is also the default for JACK internal connections, so all the plugins and jack-connected processing are likely to be the same.
Reviving a very old thread here…
So 32bit depth sounds more than alright, I’m convinced at a certain point everyone is just chasing numbers rather than using their ears
But my question is regarding the bit depth of the soundcard and how that relates to the DAW. If my soundcard is 24bit, is there any benefit in Ardour running at 32bit floating? Or worse, is there any harm in this relation?
Yes, 32bit is an established format that is directly supported by all modern CPUs and it is very efficient for arithmetic operations. This is significant for all kind of plugins or any digital signal processing.
Chances are that your soundcard already provides the 24 bit data as (zero padded) 32bit value, since busses prefer data powers of two for data transfer.
There are other benefits of converting a fixed-point value to floating point as well, and it can even be done losslessly for 24 bit data. e.g. floating point numbers have a huge dynamic range. This allows to sum many audio channels without clipping.
Definitely!
Keep in mind that the thermal noise-floor at room-temperature is at about -127dBm (@48kHz bandwidth). You get at best 18-19 bits valid data from any analog sound-source. The same during playback.
The main benefit of 24bit audio or 32bit float is convenience, not fidelity.