|
A Hopf, Skip and a Jump
|
detectorbank is a suite of modules which can be used to analyse audio files with Python, in conjuction with common modules NumPy and SciPy.
The objects it provides are: DetectorBank, DetectorCache, OnsetDetector and FrequencyShifter. Example usage of each can be found here.
This page provides advice on using these in Python.
However, due to the language differences between C++ and Python with regard to nominal vs duck typing and overloaded constructors, a few features merit further explanation.
Everything discussed below that you need to know in order to use this software is also illustrated in the Python examples.
NumPy arrays should be used for all non-scalar values passed to the DetectorBank.
OnsetDetector is given a DB
If an argument should be std::size_t, that's a Python int
A DetectorBank can be constructed either from a list of parameters or from a saved profile.
In the latter case, a profile string and input buffer must be provided (see example).
In the former, a number of parameters must be provided. The detector_characteristics argument should be a NumPy array of frequency-bandwidth pairs, which can, for example, be created from two arrays zipped together as in this example.
C++ is statically typed and therefore arguments passed from Python must be of the correct type. To prevent any errors, the sample rate, frequencies, bandwidths, damping factor and gain are cast to Python's float - and number of threads to int - when the DetectorBank is constructed.
The DetectorBank constructor also requires a Features parameter, which is explained in full here.
Both constructors take an input buffer. C++ is expecting floats; casting the input buffer to a numpy.float32 in Python may be required.
Once a DetectorBank has been constructed, you may want to use the getZ() and absZ() methods. Both take an empty array, which is filled in place (absZ also takes the getZ array). The array passed to getZ must be of type numpy.complex128
Rather than using the getZ and absZ methods directly, you can construct a DetectorCache, which will store - and discard - small buffers of absZ values, thus limiting the amount of memory used. To get a value from channel k at sample n, note that you must index the cache with [k,n], not [k][n] as you may expect.
To create a DetectorCache, you first must make a Producer, which is initialised with a DetectorBank. The Producer provides a number of methods which return information about the given DetectorBank, for example sample rate , a given channel's frequency etc.
See here for an example illustrating the above.
The FrequencyShifter is used automatically by the DetectorBank if required, but can also be used as a standalone object thus .
See here for an details on the operation of the FrequencyShifter.
1.8.13