TAPESTREA : Analysis parameters

version: 0.1.x.x (tap tap)

home: http://taps.cs.princeton.edu


Sinusoidal analysis parameters

Name Range Default Description Notes
frequency min/max 0 to 20 kHz 0 to 20 kHz defines frequency region in which to extract sinusoidal tracks  
time start/end source sound duration entire sound defines time region for extraction  
# sine tracks 1 to 50 4 number of sinusoidal peaks located per frame affects complexity of modeled sound
magnitude threshold 0.000 to 0.523 0.000 minimum peak magnitude  
threshold tilt -90.000 to 90.000 0.000 degrees by which magnitude threshold "line" can tilt across frequencies varies threshold as a function of frequency
minimum track length 0 to 20 2 minimum number of frames over which a track must continue separates longer tracks from brief peaks
frequency sensitivity 0.000 to 1.000 0.850 closeness of a track's frequency between adjacent frames, for peak matching higher sensitivity captures tracks that are stationary in frequency
allowable silence 0 to 20 0 number of frames for which a track may be dormant or invisible  
peak-to-noise ratio 0.000 to 20.000 3.100 minimum ratio of a valid peak magnitude to the average magnitude of the frame adaptive threshold local to each 10-20ms frame; lower ratios give more harmonics; higher ratios reject more noise


Grouping parameters

Name Range Default Description Notes
harmonic grouping error 0.000 to 1.000 0.100 how far a track's frequency may diverge from a group in terms of harmonics higher error means the harmonic grouping part is less strict, and tracks in a group need not be as harmonically related
common modulation error 0.000 to 2.000 0.300 how far a track's amplitude and frequency modulation may diverge from a group's higher error means it's less strict
onset error 0.000 to 1.000 0.010 maximum number of seconds between a track's and a group's onset time  
offset error 0.010 to 1.000 0.030 maximum number of seconds between a track's and a group's end times  
minimum event length 0.000 to 1.000 0.100 minimum number of seconds a group of tracks must last to be considered a deterministic event can filter out very brief single tracks that don't fit in any group
minimum track overlap 0.000 to 1.000 0.880 minimum fractional overlap between a track and a group uses overlap divided by track's or group's length


Transient detection parameters : envelope follower

Name Range Default Description Notes
attack 0.000 to 1.000 0.400 envelop follower filter's attack/ rising coefficient usually should not be too high
decay 0.000 to 1.000 0.900 filter's decay/ falling coefficient usually should not be too low
threshold 0.000 to 15.000 1.000 minimum ratio of envelope's derivative to frame's average energy, at potential transient onset higher threshold means fewer transients are found
minimum gap 1 to 22050 2000 minimum number of samples between successive transient onsets also the default transient length; increasing it includes more samples in the transient
anti-aging factor 0.000 to 1.000 0.950 weighting amount for past values, in average frame energy computation if 0, only current sample is considered; if 1, only past energy is considered; both extremes are bad


Transient detection parameters : energy ratio

Name Range Default Description Notes
long frame size 1 to 88200 22050 size of longer window, in samples  
short frame size 1 to 44100 2756 size of shorter window, in samples ratio of energy in short window to long window is compared to detect transients
threshold 0.000 to 15.000 4.500 minimum short:long energy ratio for a valid transient  
minimum gap 1 to 22050 2826 minimum number of samples between transients not equal to transient length in this case
maximum transient length 1 to 88200 44100 maximum number of samples for which a transient can last default is equivalent to 1 second at our sample rate


Raw template extraction parameters

Name Range Default Description Notes
frequency min/max 0 to 20 kHz 0 to 20 kHz defines frequency region for extraction  
time start/end source sound duration entire sound defines time region for extraction  
rolloff 0.000 to 1.000 0.200 frequency range for rolloff using a raised cosine fraction of Nyquist frequency




taps | soundlab | cs | music