Keyboard Shortcuts

MacOS Windows
play from marker space space
play selection from marker shift-space shift-space
play from start command-space ctrl-space
play selection from start command-shift-space ctrl-shift-space
play selection from start command-shift-space ctrl-shift-space

Troubleshooting Analysis Quality Problems

SPEAR tries to offer reasonable default analysis settings that should provide acceptable results in many situations. However it may be necessary to play with some analysis parameters in order to improve the quality of the resynthesis. Even with manual tweaking, sinusoidal modeling may introduce too many undesirable artifacts. Sounds which are difficult to reproduce precisely include:

Even though the reproduction of such sounds will not be faithful to the original, you may still find the results musically useful.

Also keep in mind that SPEAR does not analyze multichannel audio. If you analyze a single channel of a multichannel source, you may be missing information. Similarly if you choose to mix together a multichannel source, the analysis may have a harder time dealing with mixing of two (possibly) time incoherent sources.

You will always get the best quality resynthesis by choosing “Synthesize With Phase” or “Synthesize Without Phase” from the Sound menu. This will compute a sound file (in non-realtime) using a summing oscillator bank. If you have performed significant modifications to the analysis (such as adjusting frequencies), you should probably choose “Synthesize Without Phase” since the phases from the original analysis will probably no longer make sense.

Common Sound Quality Problems

The high end is missing — it sounds like the audio was run through a lowpass filter.
You should be able to see that high frequency information is missing from the analysis. Try decreasing the values for “Minimum Amplitude Threshold” and the “Amplitude Threshold Under Peak” in the analysis parameters window. The default value for “Minimum Amplitude Threshold” is -90. The lowest useful setting is probably around -120. The default value for “Amplitude Threshold Under Peak” is -60. The lowest useful setting is probably around -80 although it could be set as low as -90. Try decreasing both of these values gradually (in 10 dB increments) and see if the high frequency information appears.
The sound is “gritty,” especially in the low frequencies (below 100 Hz). Low frequency information appears to be poorly tracked and/or there are artifacts.
The “Frequency Resolution” parameter is probably set too high which results in too small an FFT window size. The short window (wide main lobe) is failing to resolve low frequencies or closely spaced frequencies. Try decreasing the frequency resolution value. The lowest useful setting is probably around 20 Hz.
The sound is “gritty,” even at high frequencies (above 100 Hz).
Certain kinds of sounds with a lot of dense frequency information cannot be properly tracked, even with the frequency resolution parameter reduced. This can happen with a sound that has heavy reverberation and wide vibrato for example. An acceptable result for such sounds may be impossible. The drier the sound source (the less reverberation), the better.
There are audible “chirping” noises of partials turning on and off, particularly in the high frequencies.
You should be able to see that sinusoidal tracks are stopping and starting within a certain frequency band. A single isolated chirp can be eliminated by simply deleting that partial. A series of chirps at a single frequency can be fixed by selecting the isolated partials and choosing “Join” from the edit menu to coalesce them into a single partial. If you are getting a lot of chirping (the deedly-dee effect), try decreasing the values for “Minimum Amplitude Threshold” and “Amplitude Threshold Under Peak” in the analysis parameters window (as described above).
Attack transients are unclear and time smeared.
Turn on “Perform time reassignment transient sharpening” in the analysis window (version 0.6.0 or later). Also try increasing the frequency resolution parameter as high as it will go, but without introducing gritty artifacts. Make sure that if you increase it that you are not missing low frequency information and that relevant partials are getting individually tracked. Keep in mind the general frequency content of the sound. If you are analyzing a solo soprano voice, 100 Hz would be an appropriate frequency resolution.
Stable partials are not well tracked. The frequency tracks tend to wander to spurious frequencies or to adjacent partials.
Try decreasing the “Analysis frequency tracking drift limit.” As of version 0.6.0 this is in the “Preferences” window (not in the analysis window). The default value is 0.75. The lowest useful setting is probably around 0.2.
Frequency variations such portamento or vibrato are not well tracked.
Try increasing the “Analysis frequency tracking drift limit.” As of version 0.6.0 this is in the “Preferences” window (not in the analysis window). The default value is 0.75. The highest useful setting is probably 1.0
The audio sounds time smeared, “phasey,” “reverbed,” or like it was run though a flange effect.
This is not always an easy problem to solve. The first thing to try is increasing the “Frequency Resolution” parameter as high as it will go but without introducing gritty artifacts. Make sure that if you increase it that you are not missing low frequency information and that relevant partials are getting individually tracked. Keep in mind the general frequency content of the sound. If you are analyzing a solo flute, you shouldn't need to set this to anything less than 100 Hz. If you are analyzing dense or wideband polyphonic audio, you will undoubtedly need a low value for frequency resolution (and thus a large FFT window size). Try turning on “Perform time reassignment transient sharpening” to see if the smearing is reduced. Also when resynthesizing, try choosing “Synthesize With Phase” from the Sound menu. This may reduce some of the “phasey” artifacts (this has worked will for the spoken voice in particular). Of course if you have performed a lot of modifications on the sound, synthesizing with phase information will probably not work very well.