AI Content Disclosure
Version 0.1.0-draftUpdated 24 October 2026
Draft — under legal review. Not legal advice and not yet in force.
Highlighted notes like [VERIFY: …] and blanks still to be filled in show this is a working draft.
On this page
Version 0.1 (draft) · Effective: {{EFFECTIVE_DATE}} · Last updated: 24 October 2026
In short (plain language)
- All audio made with DamruVani is AI-generated. The voices are synthetic — even when they sound real.
- We label every file: hidden information inside the file says it is AI-generated by DamruVani, with a unique ID, and we keep a fingerprint of the file so we can confirm later that it came from us.
- Free-plan audio also has a short audible tag [OWNER: keep only if the Free audible tag is live at publish], and we add an audible "AI" disclosure where the law requires it.
- You must not remove these labels, and when you publish AI audio you must say it is AI-generated wherever the law or the platform asks you to.
- Think a recording was made with DamruVani and misused? Tell us — we can check it.
1. What is AI-generated
| Feature | Output | AI-generated? |
|---|---|---|
| Text to speech, dialogue, emotion tags | Audio | Yes — synthetic speech |
| Voice clones | Audio in a cloned voice | Yes — synthetic speech that imitates a real person's voice (with that person's consent) |
| Dubbing and voice changer | Audio | Yes |
| Speech to text | Transcript / subtitles | Machine-generated text; it may contain errors |
| Auto-tag, dialect conversion, Hinglish fix | Edited text | Machine-suggested; you review it before use |
Our Library Voices are voice characters recorded by consenting adult voice actors and team members and then synthesised by our models. The character names and pictures are fictional.
Which model makes the audio: during the beta, all speech is generated by Model 1 (beta), which is built on the open OmniVoice model (CC BY-NC) and includes an audio tokenizer built with Higgs Materials licensed from Boson AI USA, Inc. and Meta Llama 3 licensed under the Meta Llama 3 Community License. Full attributions and licence texts: Credits and Licences.
2. How we label AI audio
| Layer | What it is | Where | Status |
|---|---|---|---|
| Embedded metadata | Every MP3 / WAV file carries tags saying it is AI-generated speech made by DamruVani, with a unique job identifier (ai_generated=true, dv_job_id, "AI-generated speech. generator=DamruVani") | Inside every file, on every plan — it cannot be switched off | Live |
| Provenance record | We store a fingerprint (SHA-256 hash) of every file we generate, linked to the job, account and voice that made it, for 1 year | Our systems | Live |
| On-screen label | An "AI-generated voice" badge on the Studio player, share pages and embedded players | Our apps and pages | {{ONSCREEN_LABEL_STATUS}} [OWNER: build before public beta] |
| Audible tag (Free plan) | A short, neutral sound in Free-plan audio | Free-plan audio | {{FREE_WATERMARK_STATUS}} [OWNER: not built yet] |
| Audible disclosure | A short spoken "AI voice" disclosure at the start of audio, where the law requires it | {{PREFIX_DISCLOSURE_SCOPE}} [LAWYER + OWNER: decide scope — see internal notes] | Planned |
| Inaudible watermark | A signal that cannot be heard and survives common edits, plus a public checking page | All audio | Planned (not live) |
Limits — please read: embedded metadata can be removed by re-encoding or re-recording audio, and our fingerprint matches only an exact copy of the original file. That is why we are adding an audible disclosure and an inaudible watermark, and why we ask everyone who publishes our audio to label it.
3. Why we do this
India's IT Rules, as amended in 2026, require services that let people create synthetically generated information to label lawful AI-generated content — for audio, with a prominently prefixed audio disclosure — and to embed permanent metadata or other provenance markers, including a unique identifier and information identifying the service used, to the extent technically feasible. [VERIFY: Rule 3(3) wording in the Gazette text of G.S.R. 120(E)] [LAWYER: how this applies to DamruVani's outputs] Beyond the law, honest labelling protects the people whose voices are used, our voice actors, and listeners.
4. Your responsibilities
When you use DamruVani audio:
- Do not remove or tamper with labels. Do not strip metadata, cut out audible disclosures or tags, or try to defeat watermarks (Terms §10, AUP §10).
- Disclose AI when you publish. Large social-media platforms must ask uploaders to declare synthetic content under the IT Rules — answer truthfully. Follow each platform's own AI-content labelling rules.
- Political content: cloned or real-politician voices are not allowed on DamruVani at all. Any other election-related AI audio must follow the Election Commission of India's labelling advisories. [VERIFY: current ECI advisory]
- Do not mislead. Do not present AI audio as a genuine recording of a real person or a real event in a way that could deceive listeners.
5. Checking whether audio came from DamruVani
Send the file (not just a link, if possible) to grievance@{{DOMAIN}} with your question. We check the embedded metadata and compare the file's fingerprint with our provenance records, and tell you whether it matches an output we generated. We will not disclose who created it except as the law allows (for example, to a complainant who is a victim, or to an authority under a valid order). A public checking page is planned.
6. Reporting misuse
If AI audio made with DamruVani impersonates you or harms someone, report it at /legal/grievance or with the Report button. Urgent cases are handled within hours.