Playback settings
Everything on this screen either chooses a device, chooses what leaves for the DAC, or chooses which of your designs is applied.
Input and output
The entry rate decides the family, and the family decides the internal rate: 44.1 kHz and its multiples run at 705.6 kHz, 48 kHz and its multiples at 768 kHz. The screen shows which one is open.
Output rate
A rate your DAC does not accept fails at the moment playback starts. Check what it supports in Audio MIDI Setup. A rate below the source is raised to the source rate.
Raising the output rate sends ultrasonic content to the DAC.
Anything the source or your graph holds above the old Nyquist now leaves the machine. Some DACs and amplifiers distort on it, which is why the monitor watches that band continuously and says whether the attenuation matched the design.
Output settings take effect at the next start.
The exit rate, the output format, the DSD rate and the conversion switches are saved as soon as you change them, but the device is opened with them only when the engine starts. Stop the engine and start it again to hear a change.
Leaving it at the source rate is not a compromise.
The advantage of the internal rate is that nothing folds back while the nonlinear stages run. That is already banked by the time the output stage converts once at the end.
PCM or DSD
- The modulator needs headroom to stay stable, so a fixed attenuation is applied. The amount is shown next to each choice, and it comes from the design rather than from the music.
- Turning the conversion off sends 48 kHz material out as a 48 kHz family bitstream. A DAC that does not accept that will produce high-frequency noise, so leave it on unless you know it does.
- Every choice offered stays above the analogue noise floor of a DAC, so the audible difference between orders is small. The starting values are recommendations.
- In DSD the output stage (dither, headroom and the limiter) is bypassed: what leaves is the modulator’s bitstream.
Partition length and precision
Longer partitions help heavy filters.
A long filter costs less CPU when it is convolved in longer blocks, at the cost of latency. With a short filter, lengthening the partition may not lower the load at all. The per-design validation table gives the numbers for your case.
f32 buys taps rather than quality.
At the same CPU cost, f32 carries roughly twice the taps, so the filter can be sharper. Its rounding floor is about -141 dBFS, which is the same order as the dither floor of a 24-bit output, so on a 24-bit chain the difference does not show.
Precision cannot be changed while playing.
Stop first. Both A/B slots always run at the same precision, so a comparison is never between two different arithmetic paths.
Which designs are applied
- Resample design decides how the entry rate is carried to the internal rate. Leaving it unset uses the shipped design, which is a linear phase Kaiser low pass.
- A design whose target rates do not include the current entry rate is not applied; the shipped design runs instead and the screen says so. Add the rate to the design, or write one that declares every rate.
- Graph decides what is done at the internal rate. Leaving it unset is a bypass, which is also the reference every A/B comparison is levelled against.
- Both can be changed while playing. Swapping a design in does not interrupt the sound.
- A graph containing a nonlinear node arms the output limiter for the whole session. The screen marks those graphs before you load them.