Just before World War II, American researchers started to develop a new device indented to visualize the internal structure of sounds. Because of related war interests it was given official rating as a war project, and has progressed far enough during the war period. This device firstly named "Visible speech" was brought to public attention in 1945 and was commercialized in 1951 known as Sona-Graph by Electric Kay Company. This automatic system aimed to translate sound into patterns which may be readily interpreted by the eye: these patterns called sound spectrograms are visual representations of the spectrum of frequencies of the signal as they vary with time. Inventors underlined at this time that the possible uses of this device were numerous but they did not suspect that it will make a stroke for history of music. Spectralism is an influent musical movement developed in France at the beginning of the 1970s. It is based, in part, on transcriptions of sound spectrograms. But before the emergence of this movement, François-Bernard Mâche was the first composer to resort to a Sona-Graph for composing music. In 1964, thanks to this device, he analysed the sound of his voice and transcribed the data into musical notes: Le son d'une voix for 15 instruments prefigures the spectral movement. Although its pioneer dimension, this piece was never released on recording and little is known about the creative process. How could Mâche have access to a Sona-Graph at a time where only few scientific laboratories could afford to buy such an expensive device? How did he recite an excerpt of Poésie ininterrompue II by Paul Eluard and how did he proceed for releasing the spectrograms knowing that the Sona-Graph could theoretically only analyse 2,45 seconds of sound. Which elements of the spectrograms were transcribed? Did he only translate the frequencies or did he also consider rhythmic articulations and dynamics fluctuations? To address these questions, we consulted Mâche's writings and realised that the composer never really detailed the compositional processes. He only gave in the last edition of the book Musique-Mythe-Nature a tab indicating how vowels and consonant were attributed to different instrumental sounds. Therefore we conducted a long interview with the composer at his domicile in Paris during which he explained how he conducted his spectral analysis in the Groupe de Recherche Musicale founded by Pierre Schaeffer but also in a zoology laboratory. He also detailed the creative process of this piece building on the score and an original audio recording. During our presentation, we will first remind Mâche's first artisanal attempts to transcribe voice into music in Safous Mélê (1958-59) and La Peau du silence (1962). Then, we will bring together our original data to decipher the creative process of Le Son d'une voix, underlining how the composer resorted for the first time in the history of music to a Sona-graph in order to refine his spectral analysis and open the door to the world of mimetic instrumental resynthesis.