Efficient Modeling and Synthesis of Bell-like Sounds

This page contains demonstrations related to a paper presented at the DAFx-02 conference in Hamburg, Germany, in September 2002.

Full citation:
Matti Karjalainen, Vesa Välimäki, and Paulo A. A. Esquef, "Efficient Modeling and Synthesis of Bell-like Sounds," in Proc. 2002 Conference on Digital Audio Effects (DAFx-02), pp. 181-186, Hamburg, Germany, September 26-28, 2002.

Sound Examples

This collection of sounds demonstrates how a bell sound can be resynthesized with the method that we propose in this conference paper. The synthesizer consists of a short input signal and simple digital filters: an inharmonic comb filter and a second-order resonator filter. The comb filter is made of a delay line, a second-order FIR lowpass filter, and a second-order allpass filter that are in cascade a feedback loop (for more details, see Figure 2 and Section 3 of the conference paper).

Original bell tone (423 kB, WAV format). This is a small hand-bell that we have recorded.
ARMA model of 11 lowest partials of the above tone (413 kB, WAV format). This signal is similar to the impulse response of a digital IIR filter that models all prominent partials of the bell tone and their beating up to 10 kHz. Each selected partial was modeled through the frequency-zooming ARMA technique discussed in our paper. The orders of the AR and MA parts of the model for each partial were chosen equal to 4 and 9, respectively. Modeling of the 11 lowest partials then requires a digital IIR filter with a denominator of order 44 and a numerator of order 99. To synthesize the bell sound we computed the impulse response of each partial model in the decimated domain. Then, these responses (or contributions of each partial) were mapped to the original frequency range via upsampling and de-modulation. Finally, the subband responses were added together to compose this synthetic bell sound.
Bell tone with its partials suppressed (412 kB, WAV format). The previously generated signal been subtracted from the original bell tone. This operation cancels all prominent partials up to 10 kHz. Some weak ringing still remains.
Truncated excitation signal (1 kB, WAV format). This signal contains only 300 samples, or 6.8 ms @ 44.1 kHz, taken from the beginning of the above signal. You may only hear a very short click when you play this sound file. This sample will be used as the input signal for computationally efficient bell synthesis.
Synthetic bell tone (978 kB, WAV format). Compare with the original to hear how similar/different they sound. This signal was generated by exciting the inharmonic comb filter and one second-order resonator with the short signal above. We also equalized the sound with a first-order FIR filter so that the level of the lowest partials would be close to the original ones.

Efficient Modeling and Synthesis of Bell-like Sounds

Sound Examples

Links