Projects / Switched Convolution Reverb



The Switched Convolution Reverberator

Keun Sup Lee1, Jonathan S. Abel1, Vesa Välimäki2, Tim Stilson3 and David P. Berners3


Basic Structure

Fixed Noise Sequence

Basic Structure   Fixed Noise IR

Switched Convolution Reverberator

Switched Noise IR



Switching Structures

Overlapped noise sequences
Overlapped Noise Sequences Structure
  • Pros: The most natural late-field reverb sound among the switching structures
  • Cons: The Highest computational complexity due to double length noise sequences
  • Sound Samples
    • Drum sample  Overlapped structure hihat sample (GN) Overlapped structure hihat sample (VN)4
    • Music sample  Overlapped structure music sample (GN) Overlapped structure hihat sample (VN)
  Non-overlapped noise sequences
Non-overlapped Noise Sequences Structure
  • Pros: Less complexity than the overlapped structure due to the same length noise sequences as the comb filter delay
  • Cons: Works well for single musical instruments, but for more complicated signals, such as program material, which contain both transient and steady state components
  • Sound Samples
    • Drum sample  Overlapped structure hihat sample (GN) Overlapped structure hihat sample (VN)
    • Music sample  Overlapped structure music sample (GN) Overlapped structure hihat sample (VN)
frequency dependent noise sequences
Frequency-dependent Noise Sequences Structure
  • Pros: Works better for complicated signals than the non-overlapped structure
  • Cons: Cannot use the sparse noise sequences, since each filtering operation for noise sequence spreads any pulses present
  • Sound Samples
    • Drum sample  Overlapped structure hihat sample
    • Music sample  Overlapped structure music sample
animation Continuously Varying Pulses Structure
  • Pros: Natural late-field reverb sound with low computational complexity
  • Cons: A little harsh sound
  • Sound Samples
    • Drum sample  Overlapped structure hihat sample
    • Music sample  Overlapped structure music sample

[Note]

1. For all samples, reverb times of 2 sec @ low freq. and 0.5 sec @ high freq. is used.
2. Original dry sound samples:
    - Drum sample
Overlapped structure hihat sample
    - Music sample Overlapped structure hihat sample ("Rush" from  'Cowboy Bebop OST I')
3. You can make your own samples using the SC Reverb VST plug-in which is implemented with these structures.




 

Performance Evaluation

        Complexity comparison
         Complexity comparison

        Memory requirements




Posted on Mar. 9, 2010
1 CCRMA, Stanford University
2 Dept. of Signal Processing and Acoustics, Helsinki University of Technology - TKK
3 Universal Audio, Inc.
4 'GN' = using Gaussian Noise for noise sequences, 'VN' = using Velvet Noise for noise sequences
5 All structures are implemented with a ARM926EJ-S 32-bit RISC processor.
6 The FDN (Feedback Delay Network) structure has 16 delay lines of different length, the total of which should be around 64k samples.