Description

This subroutine loads a PVOCEX file containing the PV analysis (amp-freq) of a soundfile and calculates the average magnitudes of all analysis frames of one or all audio channels. It then creates a table with these magnitudes for each PV bin.

Syntax

f # time size 43 filecod channel

Initialisation

size -- number of points in the table, power-of-two or power-of-two plus 1. GEN 43 does not make any distinction between these two sizes, but it requires the table to be at least the fftsize/2. PV bins cover the positive spectrum from 0Hz (table index 0) to the Nyquist (table index fftsize/2+1) in equal-size frequency increments (of size sr/fftsize).

filcod -- a pvocex file (which can be generated by pvanal).

channel -- audio channel number from which the magnitudes will be extracted; a 0 will average the magnitudes from all channels.

Reading stops at the end of the file.

	Note
	if p4 is positive, the table will be post-normalised. A negative p4 will cause post-normalisation to be skipped.

Examples

Here is an example of the GEN43 routine. It uses the files gen43.csd.

Note: In this example, the .csd uses the ‘system_i’ opcode to launch and run the pvanal utility and analyze a section from the sound file, finneganswake1.flac, where we find a bit of reverb and noise. The ‘reverbnoise’ selection has a duration of 0.8 seconds (-d0.8) which we find after 1.04 seconds of the source (-b1.04) and thus pvanal creates the pvocex file that we use in instrument 3 to remove this noise and reverb from the original with the pvstencil opcode.

<CsoundSynthesizer>
<CsOptions>
; Select audio/midi flags here according to platform
-odac      ;;;realtime audio out
;-iadc    ;;;uncomment -iadc if realtime audio input is needed too
; For Non-realtime ouput leave only the line below:
; -o GEN43.wav -W ;;; for file output any platform
</CsOptions>
<CsInstruments>

; by Menno Knevel 2021

sr = 44100
ksmps = 32
nchnls = 2
0dbfs  = 1

; Audacity shows the selection of reverb and noise- see image above

instr 1 ; analyze this selection and output result to pvx file
ires system_i 1,{{     
    pvanal -b1.04 -d0.8 finneganswake1.flac reverbnoise.pvx
        }}
endin

instr 2 ; untreated signal, contains some reverb and noise
asig    diskin2   "finneganswake1.flac", 1
outs    asig, asig
endin

instr 3 ; use .pvx file from instr. 1 to remove reverb and noise
; pvanal created 1 frame of size 1024, so size of table for GEN 43 = 512 (fftsize/2)
ipvx    ftgen 1, 0, 512, -43, "reverbnoise.pvx", 0    ; can be found in examples folder
asig    diskin2   "finneganswake1.flac", 1  
fsig    pvsanal   asig, 1024, 256, 1024, 1
fclean  pvstencil fsig, 0, 1, ipvx ; maximum cleaning
aclean  pvsynth   fclean
outs    aclean, aclean
endin

</CsInstruments>
<CsScore>
i1 0   2       ; first analyze
i2 5  12.7     ; untreated signal
i3 20 12.7     ; denoised gignal
e
</CsScore>
</CsoundSynthesizer>

These tables can be used as a masking table for pvstencil and pvsmaska.

f1 0 512   43  "viola.pvx" 1
f1 0 -1024 -43 "noiseprint.pvx" 0

The first example uses a 1024-point FFT phase vocoder analysis file from which the first channel is used. The second uses all channels of a 2048-point file, without post-normalisation. For noise reduction applications with pvstencil, it is easiest to skip table normalisation (negative GEN code).

Credits

Author: Victor Lazzarini