FiltAudio [options] -f FilterFile AFileI AFileO
Filter data from a single channel audio file
This program takes an audio file and a filter coefficient file as input and produces a filtered output audio file. Subsampling and interpolation factors can be optionally specified. This program supports three types of filters, FIR, all-pole, and general IIR. Filters are specified in filter files.
The first record of a filter file indicates the type of filter.
!FIR - FIR filter, direct form !IIR - IIR filter, cascade of biquad sections (5 coefficients per section) !ALL - All-pole filter, direct form
Subsequent records contain text lines with filter coefficients. Comment records ('!' in the first position of the record) can be interspersed amongst the data. Data records are free form, with data values separated by white-space (blanks, tabs and newlines). Commas can also be used to separate data values, but only within records, i.e. a comma should not appear at the end of a record.
FIR filters are specified by the direct-form coefficients h[i],
N-1 -i
H(z) = SUM h[i] z .
i=0
IIR filters are implemented as the cascade of biquadratic filter sections, where each section has a z-transform,
h(i,0)*z^2 + h(i,1)*z + h(i,2)
H(i,z) = ------------------------------ .
z^2 + h(i,3)*z + h(i,4)
All-pole filters are specified by direct-form feedback coefficients,
1 N-1 -i
H(z) = ---- , where C(z) = SUM h[i] z .
C(z) i=0
For FIR filters, a sample rate change can be affected with interpolation and subsampling. For IIR and all-pole filters, the output can be subsampled. Let Ir and Nsub be the interpolation and subsampling factors, respectively. Conceptually, the rate change is accomplished as follows.
The initial filter alignment and the number of output samples can be specified with options. The filter alignment specifies the position of the filter relative to the input data for calculating the first output sample. For FIR filters, this alignment is relative to the increased rate input sequence. Specifically, let the number of samples in the input file be Nin. The input can be considered to be an array x(0),...,x(Nin-1). The increased rate sequence is xi(.), with xi(k*Ir)=x(k). The first output sample is calculated with the beginning of the impulse response of the filter aligned with xi(idoffs). The array xi(.) can be considered to be of length Nin*Ir; the first non-zero sample is xi(0)=x(0), the last non-zero sample is xi((Nin-1)*Ir). Conceptually, the impulse impulse response is moved in steps of Nsub to create the output samples.
The intent is that the output samples are the subset of the values the would be obtained if the infinite length sequence formed by padding out the input data on either end with zeros were to be filtered. To this end, the filter calculations need warm-up points, particularly for the case that the initial filter alignment is not at the beginning of the input data. For FIR filters, this is taken into account by reading previous input values into the filter memory. For IIR and all-pole filters, previous outputs are also needed as warm-up points. If the initial alignment is near the beginning of the data, the IIR filter is run from the beginning of the data to generate the warm-up points. For larger alignment offsets, the IIR filter is backed up for a maximum of 1000 samples to provide the warm-up points.
If the initial filter alignment is not explicitly specified, it is chosen to be zero, except for symmetric or anti-symmetric FIR filters. In those cases the default alignment is Ncof/2-1 for even length filters and (Ncof-1)/2 for odd length filters. For symmetric/anti-symmetric FIR filters, this offset sets the group delay to 0 samples (odd number of coefficients) or to 1/Nsub samples (even number of coefficients).
If both the offset and the number of output samples is not explicitly set, the number of output samples is floor(Ir*Nin/Nsub). For the case of Ir=1 and Nsub=1, this results in the same number of output samples as input samples. If the filter alignment offs is explicitly set and the number of samples is not explicitly set, the number of output samples is floor((Ir*Nin-offs)/Nsub.
".au" - AU audio file ".wav" - WAVE file ".aif" - AIFF sound file ".afc" - AIFF-C sound file ".raw" - Headerless file (native byte order) ".txt" - Text audio file (with header)
"AU" or "au" - AU audio file
"WAVE" or "wave" - WAVE file. Whether or not to use the WAVE
file extensible format is automatically
determined.
"WAVE-EX" or "wave-ex" - WAVE file. Use the WAVE file extensible
format.
"WAVE-NOEX" or "wave-noex" - WAVE file; do not use the WAVE file
extensible format
"AIFF-C" or "aiff-c" - AIFF-C sound file
"AIFF-C/sowt" or "aiff-c/sowt" - AIFF-C (byte-swapped data)
"AIFF" or "aiff" - AIFF sound file
"noheader" or "noheader-native" - Headerless file (native byte order)
"noheader-swap" - Headerless file (byte swapped)
"noheader-big-endian" - Headerless file (big-endian byte order)
"noheader-little-endian" - Headerless file (little-endian byte order)
"text-audio" - Text audio file (with header)
See routine CopyAudio for a description of other options.
P. Kabal / v10r5 2023-04-10