Documentation

Run the stretcher in offline mode. In this mode the input data needs to be provided twice, once to study, which calculates a stretch profile for the audio, and once to process, which stretches it.

Run the stretcher in real-time mode. In this mode only process should be called, and the stretcher adjusts dynamically in response to the input audio.

Only meaningful in offline mode, and the default in that mode. The audio will be stretched at a variable rate, aimed at preserving the quality of transient sounds as much as possible. The timings of low activity regions between transients may be less exact than when the precise flag is set.

Although still using a variable stretch rate, the audio will be stretched so as to maintain as close as possible to a linear stretch ratio throughout. Timing may be better than when using Elastic, at slight cost to the sound quality of transients. This setting is always used when running in real-time mode.

Reset component phases at the peak of each transient (the start of a significant note or percussive event). This, the default setting, usually results in a clear-sounding output; but it is not always consistent, and may cause interruptions in stable sounds present at the same time as transient events. The Detector flags can be used to tune this to some extent.

Reset component phases at the peak of each transient, outside a frequency range typical of musical fundamental frequencies. The results may be more regular for mixed stable and percussive notes than Crisp, but with a "phasier" sound. The balance may sound very good for certain types of music and fairly bad for others.

Do not reset component phases at any point. The results will be smoother and more regular but may be less clear than with either of the other Transients flags.

Use a general-purpose transient detector which is likely to be good for most situations. This is the default.

Detect percussive transients. Note that this was the default and only option in Rubber Band versions prior to 1.5.

Use an onset detector with less of a bias toward percussive transients. This may give better results with certain material (e.g. relatively monophonic piano music).

Adjust phases when stretching in such a way as to try to retain the continuity of phase relationships between adjacent frequency bins whose phases are behaving in similar ways. This, the default setting, should give good results in most situations.

Adjust the phase in each frequency bin independently from its neighbours. This usually results in a slightly softer, phasier sound.

Permit the stretcher to determine its own threading model. Usually this means using one processing thread per audio channel in offline mode if the stretcher is able to determine that more than one CPU is available, and one thread only in realtime mode. This is the defafult.

Never use more than one thread.

Use multiple threads in any situation where Auto would do so, except omit the check for multiple CPUs and instead assume it to be true.

Use the default window size. The actual size will vary depending on other parameters. This option is expected to produce better results than the other window options in most situations.

Use a shorter window. This may result in crisper sound for audio that depends strongly on its timing qualities.

Use a longer window. This is likely to result in a smoother sound at the expense of clarity and timing.

Do not use time-domain smoothing. This is the default.

Use time-domain smoothing. This will result in a softer sound with some audible artifacts around sharp transients, but it may be appropriate for longer stretches of some instruments and can mix well with a Window setting of Short.

Apply no special formant processing. The spectral envelope will be pitch shifted as normal. This is the default.

Preserve the spectral envelope of the unshifted signal. This permits shifting the note frequency without so substantially affecting the perceived pitch profile of the voice or instrument.

Use a method with a CPU cost that is relatively moderate and predictable. This may sound less clear than HighQuality, especially for large pitch shifts. This is the default.

Use the highest quality method for pitch shifting. This method has a CPU cost approximately proportional to the required frequency shift.

Use the method that gives greatest consistency when used to create small variations in pitch around the 1.0-ratio level. Unlike the previous two options, this avoids discontinuities when moving across the 1.0 pitch scale in real-time; it also consumes more CPU than the others in the case where the pitch scale is exactly 1.0.

Each channel is processed individually, though timing is synchronised and phases are synchronised at transients (depending on the Transients setting). This gives the highest quality for the individual channels but a relative lack of stereo focus and unrealistic increase in "width". This is the default.

The first two channels (where two or more are present) are considered to be a stereo pair and are processed in mid-side format; mid and side are processed individually, with timing synchronised and phases synchronised at transients (depending on the Transients setting). This usually leads to better focus in the centre but a loss of stereo space and width. Any channels beyond the first two are processed individually.