Approved AI Voice Vendor List

Kavi Media, Inc.
Version 1.0 — Effective 20 October 2026
Published at kavi.audio/legal/ai-voice-vendors

1. About this List

1.1 This is Kavi's Approved AI Voice Vendor List (this "List"). It names every outside company that Kavi allows to make synthetic voice audio from creators' content, or to host, train or run a Voice Model or process Voice Data for Kavi. For each one it states what Kavi uses it for, what it may receive, where it processes data, and the written term that stops it training on your content. It also explains how Kavi tells you before it adds a vendor (section 4).

1.2 This List is issued by Kavi Media, Inc., a Delaware corporation with its principal place of business in Alpharetta, Georgia, USA (mailing address: 169 Madison Ave STE 30314, New York, NY 10016, USA) ("Kavi").

1.3 This List is the list referred to in: (a) the AI-Voice & Digital Replica Consent (Document J), clauses 1.12, 8.3 and 8.4 and Schedule J-1 (Part 1, "Approved Vendors"; Part 2, "Who processes it, and where"); (b) the Voice License Agreement (Document O), wherever it refers to Approved Vendors or to this List; (c) the Creator Privacy Notice, sections 5.1, 5.2(a1), 6.2 and 7.3; (d) the AI and Synthetic Content Policy, sections 6.5 and 8.3; and (e) the Contributor Privacy Notice, sections 5.1 and 6.1.

1.4 Where to find it. Kavi: (a) publishes the current List at kavi.audio/legal/ai-voice-vendors; (b) shows it in the Kavi Creator Portal ("Portal"); (c) gives a copy of the current List, with its version date and any pending additions under section 4, to every person before they sign an AI-Voice & Digital Replica Consent or a Voice License Agreement, as part of the signing packet; and (d) sends it to anyone who asks at privacy@kavi.audio, including guests, cast members, narrators and translators who do not have a Portal account.

1.5 Words used in this List. "Voice Data", "Voice Model", "Digital Replica" and "Replica Output" have the meanings given in the AI-Voice & Digital Replica Consent. "Voice Consent" means a signed AI-Voice & Digital Replica Consent or a signed Voice License Agreement that is in effect. "Approved Vendor" means a vendor listed in Part A. For a Voice License Agreement, "Voice Data" and "Voice Model" have the meanings given in that Agreement, its "Outputs" are Replica Outputs for this List, and "Kavi Voice" has the meaning given there. "Stock voice" means a ready-made synthetic voice that a vendor offers to all its customers, that is not made from, modeled on or chosen to resemble the voice of anyone in your content, and that Kavi selects itself.

1.6 Two parts. (a) Part A (section 2) is the Approved AI Voice Vendor List itself. Only vendors in Part A may make synthetic voice audio from creators' content, and only they may receive Voice Data or hold a Voice Model. Adding a vendor to Part A needs 30 days' notice (section 4). (b) Part B (section 6) lists, for transparency, the AI tool providers that process creators' content but never create, clone or alter a voice and never receive Voice Data or Voice Models. They are not Approved Vendors. They are bound by the same no-training rule (section 3.1).

2. Part A — Approved AI Voice Vendors

2.1 Approved Vendors at the date of this List

Vendor What Kavi uses it for What it may receive Where it processes data No-training term Status
A1 ElevenLabs (Eleven Labs Inc., New York, NY, USA) (a) Dubbed versions of episodes using stock voices or a Kavi Voice made under (c) (for example Hindi, Tamil, Kannada and English), as the Creator Master Agreement and section 7 of the AI and Synthetic Content Policy allow. (b) AI narration of audiobooks using stock voices (including Telugu) or a Kavi Voice made under (c), only under an AI Narration Consent in a Book Schedule. (c) Voice cloning only for a person who has signed a Voice Consent, and only for the uses that person ticked and initialed. Translated scripts and book text, for (a) and (b). Voice Data, the Voice Model and Replica Outputs only for (c). United States. ElevenLabs stores customer data in the United States by default. ElevenLabs says its affiliates, sub-processors, support staff and moderation team may also process data in other countries; it publishes its sub-processors at compliance.elevenlabs.io. Before any Voice Data goes to ElevenLabs, Kavi will add those countries to this List (section 5.2(b)). ElevenLabs' Terms of Service (section 4(i)) let a customer opt out of the use of its content for training. Once the opt-out is processed, ElevenLabs no longer uses that content to improve its services, including its models. Kavi switches this opt-out on before any creator content reaches ElevenLabs. ElevenLabs' Data Processing Addendum (section 3.1) limits its processing of personal data to Kavi's documented instructions. For voice cloning, Kavi also needs the written agreement in section 5.2(a). Approved. Not yet in use: Kavi will first use it only after the conditions in section 5 are met.
A2 Google Cloud Text-to-Speech (Google LLC, Mountain View, CA, USA) Stock voices only, for the same dubbing and audiobook narration uses as A1(a) and (b), as an alternative to ElevenLabs. Never voice cloning or custom voices. Translated scripts and book text. Never Voice Data or a Voice Model. United States. Kavi sends requests only to Google's United States (us) endpoint. Google's Cloud Data Processing Addendum allows Google to process customer data in any country where Google or its sub-processors have facilities, subject to its data-location commitments. Google Cloud's Service Specific Terms ("Training Restriction"): Google will not use customer data to train or fine-tune any AI or machine-learning model without the customer's prior permission or instruction. Google's Cloud Data Processing Addendum (section 5.2) limits processing to providing, securing and monitoring the services. Approved. Not yet in use: Kavi will first use it only after the conditions in section 5 are met.
A3 Amazon Web Services (Amazon Web Services, Inc., Seattle, WA, USA), storage only Encrypted storage, in Kavi's own AWS account, of Voice Data and Replica Outputs that Kavi holds for a Voice Consent. No AI processing of Voice Data. Voice Data and Replica Outputs, only once a Voice Consent is signed. Kavi never holds a copy of a Voice Model outside the voice vendor's systems. United States (AWS region us-east-1, Northern Virginia). AWS processes this data only to provide its storage service to Kavi, under the AWS Data Processing Addendum that forms part of the AWS Service Terms. No AWS AI service receives Voice Data. In use for hosting. Holds Voice Data only after a Voice Consent is signed.
A4 OpenAI Text-to-Speech (OpenAI, San Francisco, CA, USA), through OpenAI's API OpenAI's built-in stock voices only, for the same dubbing and audiobook narration uses as A1(a) and (b), as an alternative to ElevenLabs, in the languages Kavi uses, including Telugu, Hindi, Tamil, Kannada and English. Never voice cloning or custom voices. Translated scripts and book text. Never Voice Data or a Voice Model. United States. Kavi sends requests only from an OpenAI project set to OpenAI's United States data-residency region, which keeps storage and processing in the United States. OpenAI states that data sent to its API is not used to train or improve its models unless the customer opts in. OpenAI keeps abuse-monitoring logs for up to 30 days. Approved. Not yet in use: Kavi will first use it only after the conditions in section 5 are met.

2.1A Backup deletion. At the date of this List no vendor holds any Voice Data or Voice Model. Before a vendor first receives Voice Data or a Voice Model, Kavi adds to its row in section 2.1 the period within which that vendor deletes backup copies, as the Voice License Agreement (clauses 1.18 and 17.5(d)) requires. For a person who lives in Washington State, USA, Kavi uses only vendors whose backup copies are deleted within six months (Voice License Agreement, clause 17.5(e)).

2.2 What no Approved Vendor may do. Whatever its terms allow other customers, no Approved Vendor may, for Kavi: (a) clone, model, match or imitate the voice of any person, except the voice of a person who has signed a Voice Consent, and then only for the uses that person ticked and initialed; (b) apply a voice-preserving, voice-matching, voice-conversion or speaker-style feature to anyone's voice, except under a Voice Consent (section 3.6); or (c) receive music (masters, stems, vocals or lyrics) for any purpose.

3. How Kavi binds its vendors

3.1 No training on your content. Each vendor in Part A or Part B is bound by its written terms, together with the opt-out Kavi has exercised under them, not to train, fine-tune, improve or evaluate any model on creators' content, Voice Data, Voice Models or Replica Outputs. The only exceptions are: (a) producing a person's own Voice Model for Kavi under that person's Voice Consent; and (b) Kavi's own models that a vendor trains solely for Kavi, as section 6.5A of the AI and Synthetic Content Policy and section 7.5 of the Creator Privacy Notice describe. At the date of this List, no vendor trains any model for Kavi. If Kavi engages one, it will name it in Part B before that vendor receives any content.

3.2 No third-party training or sharing. Neither Kavi nor any vendor will use or license your Voice Data, Voice Model or Replica Outputs to train any generative model for anyone else. Kavi does not let any vendor share them with anyone, other than the vendor's own sub-processors acting for Kavi under the vendor's data processing terms.

3.3 Confidentiality and security. Each vendor must process content and personal data only to provide its service to Kavi, under written data processing and confidentiality terms. Kavi: (a) gives a vendor only the content it needs for the task; (b) uses the vendor's security settings to keep your content, Voice Data and Voice Model private to Kavi's account, with two-step sign-in and access limited to the people at Kavi who need it; (c) stores Voice Data and Voice Models encrypted, never sells, leases or trades them, and discloses them only to Approved Vendors acting for Kavi or where the law requires; and (d) does not merge any Voice Model into a multi-speaker or general-purpose model.

3.4 Deletion on request. Whenever your agreement, a Voice Consent or the law requires deletion (including on revocation, on an automatic end and on your death, as the Voice Consent describes), Kavi: (a) deletes the Voice Model and Voice Data held by each vendor, using the vendor's own deletion tools, and deletes Kavi's own copies, within 30 days (or the shorter time your Voice Consent or the law sets); (b) confirms each deletion to you in writing; and (c) treats backup copies as your Voice Consent provides: they are deleted or overwritten in the holder's normal backup cycle and are never restored or used. The backup-deletion periods are: ElevenLabs, up to 30 days after deletion; Google Cloud, within the period in its Cloud Data Processing Addendum; Amazon Web Services, Kavi's own backups, within 6 months. Kavi gives a vendor the Voice Data of a Washington resident only if that vendor's period is six months or less. Kavi also deletes working files (such as generated audio held in a vendor's history) once the finished audio is saved in Kavi's own systems.

3.5 Removal of content from vendors. When Kavi stops using a vendor, it deletes all creator content, Voice Data and Voice Models held there, as section 3.4 describes.

3.6 Stock voices or a Kavi Voice, unless you signed a Voice Consent. A dubbed or AI-narrated version uses a stock voice that Kavi chooses itself, or a Kavi Voice: a synthetic voice made only from a consenting voice actor's own new recordings under that actor's Voice License Agreement, and never modeled on anyone in the content. Kavi never uses any vendor's automatic dubbing mode that keeps, clones or matches the original speaker's voice, or any setting that picks a stock voice because it sounds like the original speaker. (For example, ElevenLabs' Dubbing feature keeps the original speakers' voices by default, and its option to turn off cloning instead picks a similar voice from its library. Kavi uses neither, except for a person's own voice under that person's Voice Consent.) Kavi uses only a vendor's own ready-made voices, never voices that other users have shared in a community voice library, and never a voice designed to imitate an identifiable person.

3.7 Security incidents. Kavi requires each Approved Vendor, under its written terms, to tell Kavi about a security incident affecting Kavi's data. Kavi notifies you by email of any security incident affecting your Voice Data or Voice Model without undue delay, and in any case within 5 business days after Kavi becomes aware of it.

4. Adding, changing or removing a vendor

4.1 30 days' notice before adding a vendor. Kavi will give at least 30 days' notice by email before it adds a vendor to Part A. Kavi sends the notice to each person who has a Voice Consent in effect, at the email address in their Portal account (or the address they gave Kavi in their Voice Consent), and also posts it in the Portal and on this page under "Pending additions". The notice states the vendor's name, what Kavi will use it for, what it may receive, where it processes data, its no-training term, and the first date Kavi may use it.

4.2 Your right to revoke during the notice period. If you revoke your Voice Consent by written notice during those 30 days, Kavi will not give that vendor any of your Voice Data or your Voice Model, and the revocation and deletion rules in your Voice Consent apply. You may revoke by email to privacy@kavi.audio, through the Portal (if you have an account), or by post to Kavi's mailing address.

4.3 No early use. Kavi does not give a pending vendor anyone's content, Voice Data or Voice Model before the end of the notice period.

4.4 Changes treated like an addition. Kavi also gives 30 days' notice under section 4.1 before an Approved Vendor: (a) receives Voice Data or a Voice Model when this List says it may not (for example, A2); (b) is used for a purpose not stated for it in Part A; or (c) processes Voice Data or Voice Models in a country not stated for it in Part A.

4.5 New signers during a notice period. If you sign a Voice Consent while a notice is running, the List you receive before signing shows the pending vendor and the first date Kavi may use it. Kavi still waits for that date.

4.6 Removing a vendor. Kavi may remove a vendor from this List at any time without notice. It then deletes content and data held there as section 3.5 describes, updates this page, and tells each person with a Voice Consent in effect by email within 30 days.

4.7 Status updates. A change of a vendor's status from "Approved. Not yet in use" to "In use", or a correction of a vendor's address or contact details, is an administrative change. Kavi makes it on this page without notice.

4.8 Part B changes. Part B providers never receive Voice Data or Voice Models, so adding one is not an addition to Part A. Kavi updates Part B, and posts a notice in the Portal, before a new provider first receives creators' content. A new AI use of your content still needs your signed agreement (AI and Synthetic Content Policy, section 11.2).

Pending additions: none at the date of this List.

5. Conditions Kavi meets before a vendor's first use

5.1 Every vendor. Before any creator content reaches a vendor in Part A or Part B, Kavi: (a) switches off every setting that would let the vendor use Kavi's content to train or improve its models, and waits for the vendor to confirm that the opt-out has taken effect where the vendor processes opt-outs manually; (b) puts the vendor's data processing terms in place for Kavi's account; (c) keeps a dated record (for example, a screenshot) of the opt-out, the settings and the terms; and (d) checks that the vendor's deletion tools work for Kavi's account.

5.2 Voice cloning. Before Kavi gives any person's Voice Data to an Approved Vendor to make a Voice Model, Kavi also: (a) has a written agreement with that vendor (such as an enterprise agreement or a signed amendment) under which the vendor will not train, fine-tune, improve or evaluate any model on Kavi's Voice Data, Voice Models or Replica Outputs (other than producing the Voice Model for Kavi), will use the Voice Model only to make audio for Kavi's account, and will delete them when Kavi asks; (b) lists in Part A every country where that vendor or its sub-processors may process the Voice Data or Voice Model, and gives notice under section 4.4(c) if that adds a country; and (c) confirms that the use is one the person ticked and initialed in their Voice Consent.

5.3 Audiobook narration. Before Kavi uses a vendor's stock voice to narrate an audiobook, Kavi confirms in writing that the vendor's terms: (a) license audiobook production and distribution of the output; (b) bar the vendor (together with Kavi's opt-out) from training on the book or the audiobook; and (c) confirm that every human voice behind the stock voice was used with informed consent and is not designed to imitate the author or any other identifiable person.

6. Part B — AI tool providers that process content but never create voices

6.1 These providers process creators' content for Kavi but never generate, clone or alter any voice, and never receive Voice Data or Voice Models. They are bound as section 3.1 describes.

Provider What Kavi uses it for What it receives Where it processes data No-training term Status
B1 Amazon Transcribe (Amazon Web Services, Inc.) Transcribing episode audio into text (Telugu, with Indian and US English where spoken), so Kavi can index episodes. Episode audio, read from Kavi's private storage. Transcripts are written to Kavi's own private storage. United States (AWS region us-east-1, Northern Virginia). Kavi has applied the AWS AI services opt-out policy to its AWS account. Under it, AWS may not use or store Kavi's content to improve its services, including for model training, evaluating service performance or debugging, and does not store it in another AWS region for that purpose. In use.
B2 Anthropic (Anthropic, PBC, San Francisco, CA, USA), Claude, through Anthropic's API Reading episode transcripts to produce keyword tags, genres and a short English description for the catalog, and an internal content-risk note for Kavi's reviewers (which is never shown to the creator and never decides anything by itself). Transcripts and episode details in text form. Never audio. Stored in the United States; processed in the United States and other countries. Kavi's Anthropic workspace stores data in the United States. Anthropic may run the processing itself (the model's work on each request) in any country where it has capacity. Anthropic's Commercial Terms of Service (section B): Anthropic may not train models on customer content from its services. Anthropic's Data Processing Addendum limits its processing of personal data to providing and maintaining its services. Anthropic deletes API inputs and outputs within 30 days, except where it must keep them longer by law or to enforce its usage policy. Kavi does not send Anthropic feedback on outputs. In use.
B3 OpenAI (San Francisco, CA, USA), through OpenAI's API Standby alternative to B2, for the same uses only (OpenAI's text-to-speech is listed separately as A4). Same as B2. Never audio. Not in use. Before first use Kavi will state the location here. OpenAI states that data sent to its API is not used to train or improve its models unless the customer opts in. Not in use. Kavi will update this List before first use.
B4 Vendors that train Kavi's own models (section 3.1(b)) None at the date of this List. — — — —

6.2 Editors' tools. Freelance editors, narrators and translators Kavi engages may use only the Approved AI Tools that Kavi lists for them in writing under their own agreements. Those tools are bound by the same no-training rule, and none of them may generate, clone or convert any voice unless that person's own agreement and a signed consent allow it.

7. Questions, requests and complaints

8. Changes to this List

8.1 Kavi changes this List only as section 4 describes. Every version shows its version number and effective date.

8.2 Every past version stays available in the Portal and on request at privacy@kavi.audio.

8.3 No change to this List can widen the uses that your signed agreement or Voice Consent allows. A new use needs a written amendment or consent signed by you.

9. Your signed agreement prevails

If this List conflicts with your signed agreement, including your AI-Voice & Digital Replica Consent or Voice License Agreement, your signed agreement prevails.

List history

Version Effective Change
1.0 20 October 2026 First List. Part A: ElevenLabs, Google Cloud Text-to-Speech, OpenAI Text-to-Speech, Amazon Web Services (storage only). Part B: Amazon Transcribe, Anthropic (in use); OpenAI (standby).

Versions

Each version is kept unchanged at its own address. The fingerprints are listed at kavi.audio/legal/policies.json.