DelayCat ReadSample Function (C++, JUCE)

Intro:

This code sample covers the inner ReadSample routine used in my advanced feature based delay line (FBDL) plugin, DelayCat. DelayCat analyzes and segments the delay buffer such that concatenative synthesis techniques can be used to create the output "delayed" signal. This routine handles reading of samples from the delay buffer, given a segment data structure which points to a segment of audio in the delay buffer and contains data associated with that segment.

This felt like a simple bite-sized section of the code that isn't too complex, yet hints at many of the larger decisions made around the implementation.

Code Overview:

The routine below reads a sample from the delay buffer, given a main output buffer [Buffer], a feedback buffer [FBuffer], a read segment data structure [Segment], and an index [i] denoting where in the buffers to write. This routine may be called multiple times for the same index [i], given different read segment structures, representing segments being read simultaneous. In this case, all calls for index [i] are summed.

This functions takes into account fractional sampling, windowing of the segment, gaining, panning, and more.

All of this logic is computed sample by sample, instead of block by block, because it may be sequentially dependent (i.e. the next sample may depend on the prior sample) with delay time < block size and feedback > 0, and I wanted this logic to be able to feed back into the buffer properly. While this does present some compromise in performance compared to the simple block by block logic, it is necessary for this application.

The read segment data structure contains data about the segment, including pointers to the front and end of the segment in the delay buffer and logic for segment operations such as reading a sample from the delay buffer, as well as control parameters for the segment. These parameters includes values used in this routine such as a segment windowing parameter, gain, pan, read speed (pitch), and feedback amount, and are computed based qualitative data (feature analysis) about the audio contained in the segment in conjunction with higher level plugin parameters.

The feedback buffer is separated from the main output buffer because the amount of feedback may be different from the audio output. This difference cannot be accounted for later in system (i.e. in the write routine) when multiple segments are summed with different feedback amounts.

A switch statement is used to handle different segment read types. Right now, only forward and backward read types are supported, so most of the logic could be factored out, but it is left as is for flexibility in adding further read types.

The temporary sample buffer [SampleBuffer] is a preallocated buffer that persists throughout the lifetime of the delay line, meaning it can be used in any operations such as this that need a temporary buffer without having to allocate any heap memory during the audio processing call, which should be avoided in realtime audio code (see my article about performant real-time audio code).

Code:

ReadSample():