Favis220c
Overview: For my 220C project, I am making vocoder renditions of choral music. The vocoder takes spectral qualities of the human voice and translates them to an instrument, creating what seems to be a "talking instrument" or "robot voice." Over the course of this project I aim to examine the effects of vibrato, pitch, vowels, and spectrum in the dry signal on the vocoded sound and discover how choral music translates to vocoded vocals.
I will be vocoding three choral pieces: Maurice Duruflé's "Ubi Caritas," Wolfgang Mozart's "Lacrimosa," and Eric Whitacre's "Lux Aurumque."
Vocoder:
I made a basic vocoder in Max MSP, but I am not satisfied with the quality of sound and annunciation of vowels/consonants. Since the goal of the project is to examine the effect of dry parameters on vocoded sound, the origin/originality of the vocoder is not necessarily a priority to me. I may end up toying with an already-made vocoder in Logic for the purposes of efficiency, timbre, and sound quality.
Methodology:
One singer is assigned to each voice for each song. Every song has bass, tenor, alto, and soprano lines, and some lines have multiple voices per line. The group will sing together during the recording sessions, but each singer will have an individual microphone placed closely to his/her mouth to isolate his/her voice from the group. The goal is to allow singers to blend during the piece while still recording the voices separately for individual vocoding.
The following singers have generously agreed to record vocals for my project:
Bass: Joel Chapman
Bass: Evan Gitterman
Tenor: Andrew Forsyth
Tenor: Will Watson
Alto: Laura Austin
Soprano: Mia Farinelli
Bass: Joel Chapman
Bass: Evan Gitterman
Tenor: Andrew Forsyth
Tenor: Will Watson
Alto: Laura Austin
Alto: Michelle Jia
Soprano: Mia Farinelli
Soprano: Grace Laboy
Bass: Joel Chapman
Tenor: Andrew Forsyth
Alto: Laura Austin
Soprano: Grace Laboy
Piano: Nayantara Jain
Rehearsals and Testing:
We held two rehearsals for each piece before recording. Both sessions went very swiftly because the singers sight-read each piece exceptionally well. During the rehearsals, I recorded some dry samples from Andrew Forsyth, the tenor voice. I had him sing phrases with and without vibrato. I also had him sing a short phrase (from Ubi Caritas) all on one note and then another take on another note. After vocoding each sample, I noticed the following:
1) Vibrato creates a nice spectral modulation in the vocoded voice. The amplitude and high frequencies flutter and add a pleasant human quality to an otherwise robotic voice.
2) When a vocoded note does not match the sung note from the dry sample, the formant of the vocoded voice is altered. Formant of the vocoded sample does not match that of the dry sample, and the vocoded sample sounds nasal as a result.
3) Consonants are well defined by the vocoder. Therefore, it is important to sample experienced singers who know how to sing consonants in traditional choral pieces (rolled r's, short s's, etc…).
Plan for Voice Vocoding:
I will go through each vocal track and play each part's appropriate notes on the keyboard vocoder. This may be difficult and tedious, but I can use MIDI in Logic to edit/split/move notes according to the rhythm of the individual voices. Choosing good reverb will also be essential for blending and authenticity.
Recording Process: CCRMA Professor Jonathan Abel has microphones that effectively isolate individual voices in a choral group. I spoke with him about using these mics and arranged a meeting with him in which he would give me the mics and run me through how to use them. This meeting was supposed to take place on the day of the recording session, but he did not show. As a consequence, we had to use a combination of condenser and dynamic mics in CCRMA studio and arrange them in a circle (in order to best isolate the individual voices). This setup was far from ideal, but we made due. All in all, the dry samples were satisfactory. However, some of the dry samples contained noticeable "bleeding," meaning that other voices spilled into the microphone that was intended to capture only one specific voice.
'Vocoding Process:'
The vocoding process itself went very swiftly. The singers' tone, vowels, and consonants were generally maintained in the vocoded product. For samples that had significant bleeding, I had to shift some formants (mainly down, to avoid nasal-sounding timbres) and adjust the formant stretch.
'Reverb:'
I knew all along that reverb was essential in simulating the timbre and dynamics of a real choral group. My first thought was to use impulse response in Logic Pro to create a customized reverb (perhaps using Stanford's Memorial Church as the venue). However, I decided to use an 8-input Faust reverb that I used in my 220B class with the help of Romain Michon. This reverb made most sense because 1) I had more control over the highs and lows and 2) the CCRMA stage has 8 inputs.
