Synthesizing Formants

The Vocoder was originally designed to improve the efficiency of voice transmission, by reducing the bandwidth necessary to maintain intelligibility. Indeed, it used a group of band pass filters to analyze the signal spectrum into a group of distinct bands, from which the original signal could be reconstructed.

A filter bank consist of a number of resonant (High Q) band pass filters, fed from a common input signal : all bands outputs can be summed or accessed independently for later processing. Obviously, filter banks can be used advantageously to simulate human vowels sounds.
For example, the Serge Equalizer can be put to good use in a patch, when synthesizing complex formants.

Needless to say, one should always take into account the specific ranges of the male/female voices:


Male voice range:
Bass : from 82,407 Hz (E2) to 261,63 HZ (C4).
Baritone: from 110 Hz (A2) to 392 Hz (G4).
Tenor : from 146,83 (D3) to 440 Hz (A4).

Female voice range:
Alto : from 196 Hz (G3) to 698,46 (F5).
Soprano: from 261,64 Hz (C4) to 1046,5 Hz (C6)


Composite Formants
--------------------------

In the 1850's, Hermann Helmholtz already studied the behaviors of resonant sounds. According to him, the french vowel U, OW, O and A could be artificially generated with a resonant tube device tuned to the fundamental frequency of the corresponding vowel.

He also found out, that two resonant tubes, tuned to two different fundamental frequencies, had to be used to generate the other five composite vowels, due to their intrinsic complex sounds characteristics (see Figure 1).







Figure 2, shows the various vowels positions, from "U to I", when turning a typical VCF frequency pot clockwise. The resonance (Q) should be set very high and the Frequency cut-off pot (Fc.) set very low.


At the bottom position, the french vowel U (as in the french word "une"), the french vowel OU (as in fool), the french vowel O (as in more), the french vowel A (as in Father), the swedish vowel Ä (as in there), the french vowel E (as in Err), the swedish vowel Ö (as in the word Öl ), the german vowel Ü (as in the german word Über) and the french vowel I (as in easy).

In the 70's, scientific researchers at the I.R.C.A.M ,in Paris, calculated that, one needs to have five formants to simulate a human vowel sound as accurately as possible.

Figure 3 and 4, shows the five typical spectral formants for the french vowel "A " sung, respectively, by a male and female performer. Notice, that the spectral curves of the male and the female formants are basically the same, but the harmonic content of the female formants exhibits a much harsher harmonics distribution.

Figure 5 and 6 shows the five spectral formants for the french vowels "O" sung, respectively, by a male and female performer. In this case, the shapes of the male and the female formants differ dramatically from each other.

Below, you will find below a table of formants for the male and female voices. Each five formants are labelled F1 to F5.

Note, that this table is three dimentional  : in X one finds the formants fundamental frequencies in Hz., in Y one finds the Amplitude in decibels (dB) but also vs. Hertz (Hz.), while in Z one finds the Amplitude vs. Hz. for each formant Bandwidth.

For example, try to plot in X/Y/Z the male formant "A", shown in Figure 3 above, using the information contained in the table below.

André Stordeur
8/25/2001