AI TEXT-TO-SPEECH COPYRIGHT CLAIMS

Legal Framework + Case Law Analysis

AI Text-to-Speech disputes usually involve three overlapping legal theories:

Copyright infringement (training data or output)

Right of publicity / voice misappropriation

Unfair competition & false endorsement

Courts rarely treat “voice” as copyrighted—but they do protect it through other doctrines.

CASE 1: Midler v. Ford Motor Co. (1988)

Facts

Ford wanted singer Bette Midler to sing a song for a commercial.

Midler refused.

Ford hired a sound-alike singer who closely imitated Midler’s distinctive voice.

The commercial never used Midler’s name—but the voice was unmistakable.

Legal Issue

Is imitating a distinctive human voice without permission actionable even if:

No copyrighted recording is used?

No name or image is used?

Holding

Yes. The court ruled in favor of Midler.

Reasoning

A person’s distinctive voice is part of their identity

Even though voices are not copyrighted, copying them can violate the right of publicity

The imitation was done for commercial advantage, which made it unlawful

Key Principle

“When a distinctive voice of a professional singer is widely known, it is protected.”

Relevance to AI TTS

If an AI voice clearly imitates a recognizable individual, it may violate publicity rights—even if:

The AI model is trained legally

The output is newly generated

This case is the backbone of celebrity voice-cloning lawsuits

CASE 2: Waits v. Frito-Lay, Inc. (1992)

Facts

Singer Tom Waits was known for his gravelly, unique voice

He refused to appear in a Doritos commercial

Frito-Lay hired a singer instructed to sound like Tom Waits

Legal Claims

Right of publicity

False endorsement

Unfair competition

Holding

Waits won and was awarded millions in damages

Court’s Reasoning

The imitation was deliberate

The voice was distinctive and identifiable

Consumers could believe Waits endorsed the product

Key Principle

“Voice misappropriation is actionable even without using the actual voice.”

Relevance to AI TTS

AI voice cloning does exactly what Frito-Lay did, but more efficiently

If an AI TTS system produces voices that:

Are clearly identifiable

Suggest endorsement or association
→ legal liability is likely

This case is cited constantly in AI voice lawsuits and cease-and-desist letters

CASE 3: Zacchini v. Scripps-Howard Broadcasting (1977)

Facts

Zacchini performed a human cannonball act

A TV station broadcast the entire performance without permission

Zacchini argued this destroyed the value of his performance

Legal Issue

Does broadcasting someone’s performance without consent violate their rights?

Holding

Yes. Zacchini won.

Court’s Reasoning

The broadcast substituted for the performance

The performer lost economic value

First Amendment did NOT override the performer’s rights

Key Principle

If reproduction replaces the original market, it’s unlawful.

Relevance to AI TTS

If AI TTS recreates narrators, audiobook voices, or performers in a way that:

Replaces hiring them

Competes directly with their work
→ this principle applies

This case is crucial when voice actors claim AI is destroying their livelihood

CASE 4: Authors Guild v. Google (2015)

Facts

Google scanned millions of copyrighted books

Created a searchable database

Displayed short text snippets

Legal Issue

Is copying copyrighted works for training/search purposes infringement?

Holding

Google’s use was fair use

Court’s Reasoning

The use was transformative

It did not replace the original books

It created a new function (searchability)

Key Principle

Transformative use weighs heavily in favor of fair use.

Relevance to AI TTS

This case is used by AI companies to argue:

Training on text data is fair use

Training is different from output

BUT:

This protection weakens if the AI output reconstructs expressive elements, including voice style

Courts distinguish training from output exploitation

CASE 5: Andersen v. Stability AI (2023–ongoing)

Facts

Artists sued AI companies for training models on copyrighted works

Claimed unauthorized copying and derivative outputs

Legal Questions

Is training itself infringement?

Are AI outputs infringing derivatives?

Early Court Findings

Training alone may be allowed

But outputs that resemble specific artists may create liability

Plaintiffs must show substantial similarity

Key Principle

AI systems are not immune; outputs matter.

Relevance to AI TTS

TTS models trained on copyrighted audiobooks or performances:

Training may be lawful

Output that mimics specific voices may not be

Courts increasingly focus on what the user hears

CASE 6: Getty Images v. Stability AI (2023–ongoing)

Facts

Getty sued Stability AI for training on proprietary images

Evidence showed outputs containing Getty watermarks

Legal Significance

Demonstrates courts care about training data provenance

Outputs revealing source material weaken fair-use claims

Relevance to AI TTS

If TTS outputs:

Reproduce cadence, phrasing, or vocal traits traceable to a dataset

Reveal proprietary or licensed voices
→ risk increases significantly

CASE 7: In re: TikTok Text-to-Speech Voice Litigation (2023)

Facts

Voice actor alleged TikTok used her recorded voice to build TTS

Claimed lack of consent and misappropriation

Legal Theories

Right of publicity

Unjust enrichment

Breach of contract

Importance

One of the first modern AI TTS voice lawsuits

Shows shift from theoretical risk to real litigation

CORE LEGAL TAKEAWAYS FOR AI TTS

1. Voices Are Not Copyrighted—but They Are Protected

Protection arises through:

Right of publicity

Unfair competition

False endorsement

2. Training ≠ Output

Courts increasingly separate:

Training legality

Output liability

3. Distinctiveness Is Key

The more recognizable a voice is:

The higher the legal risk

Especially for commercial use

4. Consent and Licensing Matter

Strong defenses include:

Explicit voice licenses

Synthetic or blended voices

Clear disclaimers

FINAL SUMMARY

AI Text-to-Speech copyright disputes are not a legal vacuum. Courts already have:

Tools

Precedents

Doctrines

They simply apply them to new technology.