ftconv

Low latency multichannel convolution, using a function table as impulse response source.

The algorithm is to split the impulse response to partitions of length determined by the iplen parameter, and delay and mix partitions so that the original, full length impulse response is reconstructed without gaps. The output delay (latency) is iplen samples, and does not depend on the control rate, unlike in the case of other convolve opcodes.

Syntax

a1[, a2[, a3[, ... a8]]] ftconv ain, ift, iplen[, iskipsamples \
                                [, iirlen[, iskipinit]]]

Initialization

ift -- source ftable number. The table is expected to contain interleaved multichannel audio data, with the number of channels equal to the number of output variables (a1, a2, etc.). An interleaved table can be created from a set of mono tables with GEN52.

iplen -- length of impulse response partitions, in sample frames; must be an integer power of two. Lower settings allow for shorter output delay, but will increase CPU usage.

iskipsamples (optional, defaults to zero) -- number of sample frames to skip at the beginning of the table. Useful for reverb responses that have some amount of initial delay. If this delay is not less than iplen samples, then setting iskipsamples to the same value as iplen will eliminate any additional latency by ftconv.

iirlen (optional) -- total length of impulse response, in sample frames. The default is to use all table data (not including the guard point).

iskipinit (optional, defaults to zero) -- if set to any non-zero value, skip initialization whenever possible without causing an error.

Performance

ain -- input signal.

a1 ... a8 -- output signal(s).

Example

Here is an example of the ftconv opcode. It uses the file ftconv.csd.

<CsoundSynthesizer>
<CsOptions>
; Select audio/midi flags here according to platform
; Audio out   Audio in
-odac           -iadc    ;;;RT audio I/O
; For Non-realtime ouput leave only the line below:
; -o ftconv.wav -W ;;; for file output any platform
</CsOptions>
<CsInstruments>
sr      =  48000
ksmps   =  32
nchnls  =  2
0dbfs   =  1

garvb   init 0
gaW     init 0
gaX     init 0
gaY     init 0

itmp    ftgen   1, 0, 64, -2, 2, 40, -1, -1, -1, 123,           \
               1, 13.000, 0.05, 0.85, 20000.0, 0.0, 0.50, 2,   \
               1,  2.000, 0.05, 0.85, 20000.0, 0.0, 0.25, 2,   \
               1, 16.000, 0.05, 0.85, 20000.0, 0.0, 0.35, 2,   \
               1,  9.000, 0.05, 0.85, 20000.0, 0.0, 0.35, 2,   \
               1, 12.000, 0.05, 0.85, 20000.0, 0.0, 0.35, 2,   \
               1,  8.000, 0.05, 0.85, 20000.0, 0.0, 0.35, 2

itmp    ftgen 2, 0, 262144, -2, 0
       spat3dt 2, -0.2, 1, 0, 1, 1, 2, 0.005

itmp    ftgen 3, 0, 262144, -52, 3, 2, 0, 4, 2, 1, 4, 2, 2, 4

       instr 1

a1      vco2 1, 440, 10
kfrq    port 100, 0.008, 20000
a1      butterlp a1, kfrq
a2      linseg 0, 0.003, 1, 0.01, 0.7, 0.005, 0, 1, 0
a1      =  a1 * a2 * 2
       denorm a1
       vincr garvb, a1
aw, ax, ay, az  spat3di a1, p4, p5, p6, 1, 1, 2
       vincr gaW, aw
       vincr gaX, ax
       vincr gaY, ay

       endin

       instr 2

       denorm garvb
; skip as many samples as possible without truncating the IR
arW, arX, arY   ftconv garvb, 3, 2048, 2048, (65536 - 2048)
aW      =  gaW + arW
aX      =  gaX + arX
aY      =  gaY + arY
garvb   =  0
gaW     =  0
gaX     =  0
gaY     =  0

aWre, aWim      hilbert aW
aXre, aXim      hilbert aX
aYre, aYim      hilbert aY
aWXr    =  0.0928*aXre + 0.4699*aWre
aWXiYr  =  0.2550*aXim - 0.1710*aWim + 0.3277*aYre
aL      =  aWXr + aWXiYr
aR      =  aWXr - aWXiYr

       outs aL, aR

       endin

</CsInstruments>
<CsScore>

i 1 0 0.5  0.0  2.0 -0.8
i 1 1 0.5  1.4  1.4 -0.6
i 1 2 0.5  2.0  0.0 -0.4
i 1 3 0.5  1.4 -1.4 -0.2
i 1 4 0.5  0.0 -2.0  0.0
i 1 5 0.5 -1.4 -1.4  0.2
i 1 6 0.5 -2.0  0.0  0.4
i 1 7 0.5 -1.4  1.4  0.6
i 1 8 0.5  0.0  2.0  0.8
i 2 0 10
e

</CsScore>
</CsoundSynthesizer>

See also

Convolution and Morphing

Credits

Author: Istvan Varga
2005