Voice Rights & AI Disclosure

Last updated: 2026-10-09

Iconic Voices publishes audiobook narrations of texts that are in the public domain in the United States. Every narration we publish is AI-generated audio: an AI restoration where recordings of the author's real voice survive, a studio-designed AI voice where they do not — never a lost archival recording and never passed off as one. No human actor performs on this catalog. The author portraits you see next to the voices are AI-generated images, not photographs. Every book is labeled before you press play. This page is our transparency disclosure under EU AI Act Article 50 (in force since 2 August 2026) and explains exactly what each label means — and, below, what we can and cannot say about the recordings the restorations are built from.

The three labels

Every voice carries exactly one of these three labels. All three describe AI voices.

• Restored AI voice — AI voice restored from real recordings. A voice model built from surviving recordings of that person's actual speaking voice: not an actor, not a descendant, not a portrayal. What the model learns is timbre, cadence, the shape of the vowels. What you then hear is newly generated speech — the author's public-domain text, read in a reconstruction of how the author sounded. It is not a recovered recording of them reading the book. No such recording exists for any book in this library.
  
  How much survives varies enormously. For some authors there are hours of broadcast talks and studio sessions. For others there are a few minutes, or it is not documented which recording the model was trained on. Where the surviving material is that thin, the model is inventing far more than it is restoring, and the voice page says so: it adds AI voice · limited archival source and tells you that much of what you hear is AI synthesis rather than restoration.
• Studio AI voice — a new AI performance. Where no authentic recording of the author survives (Darwin, Tolstoy, Adam Smith and many others), no restoration is possible. For those books the narration is performed by an AI voice designed in our studio for that author: a historically informed interpretation, not a reconstruction. It is not built from recordings of the author, it is not the author's voice, and it is not a human narrator. If we cannot show that a recording of an author's voice survives, that author belongs here, not under the restored label.
• AI voice · disputed source — an AI voice modeled on archival material that cannot carry a full restoration: the surviving recording is only a few lines or minutes long, it is in a different language from the book, or it is not settled which recording, or whose voice, the model rests on. The voice is still a reconstruction attempt, but the source is open to question, so we label it that way rather than as restored.

The label for each book appears on its voice page, in the player, and in the iPhone app — always before you press play. Each author's voice page carries a provenance badge stating which label applies; that per-page badge is the authoritative answer to "who narrated this book?".

AI-generated portraits

The portraits of authors on this website, in the iPhone app and in link previews are AI-generated images. They are not photographs, they do not show real scenes or settings, and some are deliberately photorealistic. Wherever one appears it is labeled as AI-generated — "AI-generated portrait" on the website and in link previews, "AI portrait" in the app — and its text alternative for screen readers says the same.

• They are artistic interpretations informed by public historical references, made so the library has a face for each voice. They are not a record of how the person looked on any given day.
• A portrait, like a voice, does not mean that the person, their estate or any archive endorses Iconic Voices.
• Some US states protect a deceased person's likeness as well as their voice (see below). The removal process on this page covers portraits exactly as it covers voices.

Where the recordings come from — and what we can honestly say about their status

Every text we narrate is in the public domain in the United States; availability varies by country. In the US, books published in 1930 or earlier are in the public domain, and that is why we can publish them there. Canada, the United Kingdom and Australia count copyright from the author's death instead, so a few texts that are free to use in the US are still protected in one of those countries, and we restrict those books and samples there.

The recordings are a separate question with a separate answer, and we used to blur the two. US law does not give an old recording the same term as an old book. Sound recordings fixed before 15 February 1972 run on their own schedule (the Music Modernization Act, 17 U.S.C. §1401): those first published before 1923 entered the public domain on 1 January 2022; recordings published between 1923 and 1946 get 100 years from publication; 1947 to 1956 get 110 years; and anything else fixed before February 1972 stays protected until 15 February 2067. A recording made in the 1930s or 1940s is therefore not in the public domain today, and will not be for years or decades. Film and newsreel soundtracks follow a different rule again, and a recording that is out of copyright abroad can still be protected here.

In 2026 we audited every restored voice against those rules, author by author, with sources. The honest summary:

• A small number of restorations rest on recordings that appear to be in the US public domain — mostly commercially issued discs and cylinders from before 1923.
• Most rest on recordings that are still under copyright, or held by an archive under access terms, or both.
• For a few, the record is genuinely unresolved: whether the recording was ever formally published, or which transfer survives, cannot be established from the available evidence.

So we no longer describe the sources as public-domain archival recordings, because for most of them that is not true. Instead, each voice page names its own source — the recording, its date and occasion, the institution that holds it, and what we can and cannot establish about its status. Where the status is unresolved, the page says that too.

We are also not claiming the opposite. Naming a source is not a claim that we hold a license to it, and it is not a claim that anyone else's rights have expired. It is a statement of what the recording is, so that anyone with an interest in it can see exactly what we used and tell us to stop.

One more thing that belongs in the open. Several US states protect a deceased person's voice and likeness for decades after death — California for 70 years, New York for 40 (for deaths since May 2021), Tennessee under the ELVIS Act, Indiana and Oklahoma for 100 years, Hawaii for 70. For a number of the authors in this library those windows are still open. That is one of the reasons the removal process below is fast, unconditional, and does not require anyone to prove a case to us first.

Are these the authors' real voices?

No. You are never hearing a lost archival recording of the author reading the book. Where recordings of the author's voice survive, we use them to build the restored voice model, and the result is labeled Restored AI voice. Where nothing usable survives, the book is labeled Studio AI voice. Where the surviving source is too thin, in another language, or contested, the label says AI voice · disputed source.

How the disclosure is actually made, today

We would rather be exact than flattering, so here is what exists and what does not. Every book carries its label in the listening interfaces — the voice page, the player, and the iPhone app — always before you press play; every portrait carries its "AI-generated portrait" label; and this page carries the disclosure in machine-readable page metadata. What we have not shipped yet is a marking inside the audio files themselves: an MP3 downloaded from us contains no embedded tag or watermark saying it is synthetic, so a file separated from the page it came with no longer carries its own disclosure. We are working on embedding that marking; until it ships, treat this paragraph as the honest limit of the claim. The EU AI Act's Article 50 transparency obligations for synthetic audio have been in force since 2 August 2026, and this page and the per-book labels are how we meet the disclosure duty in the meantime.

Can I request that a voice or a portrait be removed?

Yes — the default response is removal. Email legal@iconicvoices.io with the subject "Voice removal request", identifying:

• Your relationship to the author, speaker, or recording — estate executor, heir, performer, representative, or the archive, broadcaster, publisher or label that holds the source recording
• The specific content you would like removed (author, title, portrait, ad creative, URL if on our site)

We acknowledge within 24 hours and complete removal within 24 to 48 hours. We do not require a formal DMCA notice and we do not litigate takedowns. This applies to rights in the source recording just as much as to rights in a person's voice or likeness — if you hold the recording we built a voice from and you want it gone, that is enough.

Why we do this

Historical knowledge locked in public-domain books deserves a wider audience. A restored voice lets modern listeners hear a text the way its author sounded; an honest narration keeps the rest of the library alive. We chose maximum transparency as the price of doing this at all: every text we narrate is in the public domain in the United States, every voice and every portrait is labeled as AI, every source is named on its own page, and every takedown is respected.

Standards and rules we work to

• EU AI Act, Regulation (EU) 2024/1689, Article 50 — transparency obligations for synthetic audio and images. In force since 2 August 2026.
• 17 U.S.C. §1401 (Music Modernization Act, 2018) — the term schedule for pre-1972 sound recordings, described above.
• 17 U.S.C. §101 — the definition that keeps film and television soundtracks out of the sound-recording schedule and on the ordinary term for films.
• California Civil Code §3344.1, as amended by AB 1836 (2024) — name, voice, likeness and digital replicas of deceased persons (70-year post-mortem period).
• California AB 2602 (2024) — digital-replica rights of living performers.
• New York Civil Rights Law §50-f — digital replicas of deceased persons (40-year post-mortem period, for deaths on or after 29 May 2021).
• Tennessee ELVIS Act, Tenn. Code Ann. §47-25-1101 et seq. (effective 1 July 2024) — voice and likeness protection.
• Indiana Code §32-36-1 and Okla. Stat. tit. 12 §1448 — rights of publicity that expressly cover voice and run 100 years after death.
• Hawaii Revised Statutes ch. 482P — a property right in a person's voice and likeness, 70 years after death.
• Archive and deposit terms attached to individual recordings, which we treat as binding whether or not copyright has expired.

Questions or removal requests

Voice, portrait and rights questions: legal@iconicvoices.io. General legal questions: legal@iconicvoices.io. Everything else: Support.