Tagh o Tagh · AI voice imitation
AI imitation of Mahasti — not Mahasti
An explicitly AI-generated imitation, with a discussion of voice, artistry and consent.
The audio file is included for you to save and keep. Please retain its AI label when sharing or saving it.
Original singer and composer: Mojtaba Dorbidi Lyricist: Mani Armaghan Original arranger: Hamed Abdollahi Today's version: an AI imitation of Mahasti's voice, not a performance by Mahasti herself.
Why today?
After comparing performances by different singers, today we turn to a different question: does every performance that resembles a singer's voice also reflect their artistry and choices?
A look at the work
Tagh o Tagh is originally the song Oomadam, released by Mojtaba Dorbidi in March 2023. It is not a newly discovered Mahasti recording. The lyrics describe arriving with gifts and preparations intended to delight a loved one.
Musically, short, recurring phrases give the main section a rhythmic character, and the pronunciation of the opening words works almost like percussion. Repetition also helps the song stay in the listener's memory and invites participation. What makes this version striking is the meeting of this direct, contemporary pop expression with a voice intended to remind us of Mahasti.
Here, timbre matters: the quality that allows us to recognize a voice independently of the melody it sings.
These covers have their roots in research into generating and converting human voices. In one common approach, a model learns a singer's vocal characteristics from existing recordings. It then transforms another performance to resemble that singer while preserving the words, melody and much of the original timing. The result is newly generated sound, rather than pieces of old recordings joined together word by word. Accessible software has also made this possible outside professional studios.
The distinction matters musically. The vocal colour may recall Mahasti, while the phrasing and delivery still come from someone else. Imitating her voice does not tell us which notes she would have held, where she would have paused, or whether she would have agreed to sing the song at all.
For someone who grew up with Mahasti's voice, hearing that resemblance in a new song can be deeply enjoyable. It might even send them back to her real performances. But the pleasure of listening does not, by itself, settle the question of consent.
An artist who is no longer alive cannot reject a lyric, request another arrangement or object to the use of their name. An AI label helps with transparency, but does not itself replace an artist's consent.
For those who read these reflections to the end: At what point does keeping a singer's memory alive become making decisions on their behalf? Is disclosing that a performance is AI-generated enough to make it a tribute, or is the artist's prior consent also necessary? Would your answer change if their family approved, or if its creator earned money from it?
More from the archive
View allMorgh-e Sahar
Shajarian, Bahar and Neydavoud: the voice of grief and persistence.
Aseman-e Aabi × Love in Portofino
One melody, two languages, and two very different stories of love.
Sarab-e Ghalbam
Hayedeh’s later musical language and a recording left unfinished.
