This thesis proposed the use of modulation to reduce the memory and processing requirements of artificial reverberation algorithms.  Several modulation sources have been evaluated and the sinusoidal tone generator has been retained as one of the most efficient and was found appropriate for use as a modulator.  Values of parameters such as modulation depth and rate have been suggested to achieve a good sounding reverberation, although they could be set differently according to the type of input.  Different interpolation methods have also been reviewed.  The modified maximally flat group delay design method to design an all-pass IIR interpolator has been retained as the most efficient and sonically transparent interpolation type.  Several implementation methods and optimizations were then proposed to make the modulated algorithm more efficient.  This led to the design of a reverberator using 4 modulated delay lines out of 8.  This time-varying design performs slightly faster and requires only 7k words; about 60 % of the memory of the conventional design using 12 non-modulated delay lines.  A listening test proved that this design sounded better than the conventional design on three recordings out of four.

Improvements could still be made to the design by trying other modulation types or more efficient random number generator modules such as shift register RNGs.  Coefficients of the feedback matrix could also be modulated, instead of (or in addition to) the feedback delay lengths.  The output matrix could also be modulated to change in real-time the proportion of each delay line in the output.

A typical implementation of this time-variant algorithm would include additional early reflections during the first 70 ms of reverberation.  On a 16-bit gaming platform with a sampling rate of 48 kHz, the complete algorithm would take about 20 kBytes of RAM.  As we already mentioned, the built-in reverberation algorithm on a PlayStation I takes up to 100 kBytes of RAM.  Using our time-varying algorithm would thus result in a maximum saving of about 80 kBytes of memory.  This represent a gain of almost 8% of the total RAM (1 Mbytes) allocated for all audio files and effects.

Other applications with limited memory resources would benefit as well from the use of modulation in their reverberation algorithms.  For example, most hardware synthesizers and sound cards [6] with built-in digital signal processor (DSP) have limited RAM to execute reverberation algorithms.  In fact, most inexpensive DSP only have an internal RAM of 1 to 10 kBytes.  Some virtual reality applications [18] [19] [22] [37], multi-media applications [20] [21], and computer music applications [32] are also extremely demanding in term of performance and have very limited memory resources.  Finally, we should mention again that high-end reverberation units could benefit as well from the use of modulation to enhance the quality of their algorithms.