Farinelli, or a comment on recreating the voice of a castrato by fusion of timbres

Carlo Broschi, called Farinelli (1705-82), is known as one of the most famous castrati. Obviously, we have no recording of his unique voice. For the needs of the movie of Gerard Corbiau, devoted to this singer, Stephan Films and Auvidis have both approached the IRCAM to recreate an unheard voice, quite beyond the reach of modern human performance, using digital techniques of sound processing.

Castration was banned in the 19th century, and the last castrato in the Western World died in 1922. Castrati were virtuoso musicians, exceptionally talented and trained. Almost nothing in their repertoire can be performed nowadays.

Castrati were particularly known for their unique timbre: because of the surgery performed on them, their voice did not change with puberty. Upon adulthood, the size of their thoracic cage, their lung capacity, their physical stamina and their strength were usually above that of most men. They had, as a consequence, great vocal power, and some were able to sing notes for a minute or more. Finally, a small and flexible larynx, and relatively short vocal chords allowed them to vocalize over a rather wide range (over 3 and 1/2 octaves) and to sing with great agility (they could control wide intervals, long cascades and trills). Furthermore, castrati were initially selected among the best singers and received intensive training.

The greatest difficulty in making a movie on a castrato is the absence of recorded material. The last castrato in the Western world, Alessandro Moreschi, recorded less than one hour's worth of singing on wax cylinders between 1902 and 1904. The technical quality of this historical recording is such that it is impossible to obtain acoustic data from it, and it serves only as an indicator of the general aesthetic qualities of the voice.

So as to recreate a voice true to the original, we started from the notion that we had to call on performers of today, and, with the help of techniques developed recently, transform their natural voices so as to obtain a timbre close to that of a castrato.

To better identify this lost voice and to define its characteristics, we have taken into account the physical traits of the organs involved in voice production in a castrato, the general aesthetic qualities of the available historical recording, and of descriptions of the singing found in written accounts. The movie director and the musical adviser to the movie were also involved in the artistic choices.

We have attempted several approaches, starting from a bass, a tenor, an alto, a countertenor and a coloratura soprano. Because of the very wide vocal range, we have decided to use both a countertenor (Derek Lee Ragin) and a coloratura (Ewa Godlewska) who used similar singing techniques (especially with respect to vibrato and articulation).

After the selection of the singers, the processing took place in two stages. The first, at the "Image et son" in Neufchatel, Switzerland, carried out by Jean-Claude Gaberel, tried to recreate the melodic line of the castrato voice based on recordings of the two singers. This was done by splicing the parts which could only be sung by the countertenor with those specific to the soprano. This subtle editing work was sometimes done note by note.

The second stage was carried out by the analysis-synthesis team of the IRCAM: the aim was to blend the timbre of the two voices. A timbre close to that of the countertenor, itself quite characteristic, was chosen as reference. This voice was nevertheless processed to give it a younger quality, in particular removing certain noisy aspects. The voice of the soprano was then modified more dramatically by transforming it toward that of the countertenor. Additional processing was required to produce certain effects such as very long notes which could not be sustained by today's singers. These notes were wholly synthesized with elements sampled from other parts of the sung material.

So as to better understand the type of transformation that was applied, an analogy with image processing may be useful: it is now possible to create sequences which represent a continuous transformation from one scene to another. One of the most common applications alters the face of one person into that of another This procedure, called interpolation of forms or morphing, defines a structure of the scenes, based on position, shape, texture, outline of each object present in the scene. Morphing is a gradual transformation from one scene into the other by continuously changing the shape, texture and outline of each fundamental element of the face. Likewise, we go from one voice to the other by gradually altering the characteristics of one voice into those of the other voice.

To implement these subtle transformations, we have used programs developed as part of our research at the IRCAM and developed other programs specially for this project. In particular, each vowel was analyzed, identified in terms of its nature, pitch and amplitude, and transformed in a particular way. Graphical tools were used to compare analyses and modify them in real time for testing purposes. A special synthesizing process, known as FFT-1 and patented by the IRCAM, was used to create long vowels.

The digitalized sound was processed on one of the most powerful workstations, a Dec Alpha600, provided by DEC France for the purposes of this project. In all, 45 minutes of vocal material were transformed to produce the soundtrack of the movie, thus allowing to bring back to life some of the repertoire.

To Early Music FAQ

translated by François Velde from original article by Sony France