Temporada 2 · Cartas desde el Polvo Marciano · JVV-T2-23

Necromantic algorithms: resurrecting voices with necro-phonic GANs

I Talking to the echoes

In 2024 a study at the University of Helsinki managed to synthesise 30 seconds of Sibelius's voice from six scratched wax cylinders.
Today, DiffWave-GAN models train on barely three minutes of damaged audio and rebuild the timbre, prosody and breathing of any dead person with a fidelity that raises the hairs on your skin.
The Churches of the Void sing psalms with extinct choirs; memorial unions recreate the statements of murdered workers; AAA studios plan “posthumous” cameos in CGI sagas.

Necromancy no longer summons spirits: it compiles datasets.

II Pipes of the macabre miracle

Stage	Tech 2026	Latent danger
Excavation	A spectral restoration AI cleans magnetic tapes	It can uncover private conversations never meant to be public
Embalming	Neuro-Vocoder analyses the formants; builds a latent model < 5 MB	Any leak = a perfect deep-fake
Possession	Prompt-TTS combines text with an emotion parameter {–1, +1}	The living actor is paid less, SAG-AI protests
Transmission	Live STT API, latency < 120 ms; streamed into a podcast, a call centre or an XR altar	Identity theft in “desperate grandchild” frauds

III Emerging necro-phonic liturgies

Mass of the Lost Voice – Cathedrals project sermons by 16th-century nuns rebuilt from phonetic parchments.
XR workers' council – Holograms of miners from 1890 describe working conditions to ILO delegates.
Posthumous Album DAO – Fans crowdfund a new album by a dead singer; the royalties are shared with foundations chosen by her family.

(None of them includes sexually explicit or violent content.)

IV Current ethical-legal limits

Jurisdiction	Status 2025	Pending key issue
EU	The “Deepfake Act” demands an audible watermark every 15 s if the person did not consent in life	Certifying the authenticity of consent
California	Post-mortem right of publicity, 70 years; the estate decides	Synthesised voiceprints count as a taxable “performance”
Mexico	No clear framework; cases of narcocorridos with dead voices are in the courts	Creating a national register of heritage voices

V Technical safeguards

Inaudible watermark
– A –60 dB phase-noise pattern; public detectors confirm the synthesis.
“Vox-License” contract
– A soul-bound NFT that specifies: duration, purpose, % of royalties, the family's right of revocation.
Self-expiring model
– The latent weight encrypts a key that expires in 10 years unless the heirs renew it.
Sonic civil register
– A decentralised library (IPFS + ledger) stores real samples for forensic comparison.

VI Futures (2030)

Path	Result
Just memory	Voices resurrected only with consent; auditory heritage democratised.
Grey market	Tiktokers use speeches by every famous dead person; historical banalisation.
Spectral censorship	States ban any synthesis of leaders; vocal smuggling begins.

VII Coda

“Resurrecting the voice is resurrecting the responsibility.”

If every whistle of the past can re-sound from a prompt, we need rites and laws that honour the dead, protect the living and stop the echo becoming a weapon.
Between nostalgia and fraud, necro-phony will find its balance in transparency, a clear licence and the ability to fall silent when silence is still the better memory.