A Virtually Real Flute

Since the first keyboard-controlled digital synthesizers became available, several new synthesis interfaces have been developed (e.g., Mathews 1991a, 1991b; Cook 1992; De Laubier 1998). As most of these digital instruments differ considerably from traditional instruments, musicians must learn new techniques to play them (Kronland-Martinet, Voinier, and Guillemain 1997). Here, we propose overcoming this difficulty by designing a digital flute using a traditional instrument form factor to control a synthesis model. The digital flute was assumed to extend the technical scope of the traditional flute, but we also wanted to be able to use the instrument in the traditional way. To connect the instrument to a computer, we added sensors to its key pads and placed a microphone inside the mouthpiece. The synthesis model to be controlled by this interface had to take the physical characteristics of the instrument into account. A physical model was therefore developed to simulate the propagation of waves inside the flute.

The system of excitation involved in flute-playing is highly complex from a physical point of view. To construct a real-time model with parameters that can be measured while the instrument is being played, we used a signal model to simulate the source excitation. By injecting this model into the physical one, we constructed a hybrid model which accounts for both the physical and perceptual aspects of the sound produced.

Design of the Interface

Playing a wind instrument involves two main factors. The first of these is the player's finger position, which is correlated with an equivalent length of the instrument (Nederveen 1998) and thus with the pitch, and the way the instrument is excited by the air jet (Fletcher and Rossing 1990; Verge 1995). This information must be detected and combined in order to control a synthesis model in real time or to produce standard MIDI messages.

Pitch Detection

The pitch is determined by both the player's finger position and the way the instrument is blown. Signal processing methods can be used to analyze the sound emitted by the instrument and accurately estimate the pitch. Since the flute is a monophonic instrument, a pitch extractor of this kind can be used to directly perform the MIDI encoding of musical sounds without having to solve the problems associated with polyphonic instruments. The fiddle~ MSP object (Puckette, Apel, and Zicarelli 1998) is a good example of an available tool which is well suited to applications of this kind. In this case, the instrument only needs to be equipped with a microphone connected to the sound input of a computer running an MSP program.

In our case, we wanted to be able to control the synthesis model with the real instrument, even when the flute is not blown. The state of the key pads therefore must be detected to obtain information about the player's finger position. In addition, the key pad noise is of musical relevance, and we therefore had to collect information of another type: the speed at which the key is pressed. To detect the time-varying position of the keys, we used a combination of magnets and Hall effect sensors. A Hall effect sensor gives an output voltage which is a function of the magnetic field received. If the magnetic field is generated by a permanent magnet, its intensity will depend on the square of the distance between the sensor and the magnet. The output voltage of the sensors is then correlated with the spatial distance between the key pads and the corresponding holes, which makes it possible to detect whether the hole is opened or closed. By regularly sampling this voltage, we can estimate the speed at which each key is pressed or released. For practical reasons, the magnets were connected to each finger key on the instrument, while the Hall effect sensors were placed in front of each magnet on an aluminum rail placed parallel to the instrument, as shown in Figure 1.

The magnetic...