Frequently asked questions
Do I need consent to use preset voices like Aria or Nova?
No. The voices are based on licensed recordings from LibriTTS-R (CC BY 4.0) and LibriVox public-domain audiobook readings. You do not need individual consent from the voice actors to use these presets in your generated audio files.
What is the difference between voice cloning and preset voices?
Preset voices are eight fixed options (e.g., Aria, Nova) derived from existing recordings. Voice cloning involves creating a unique digital replica of a specific person’s voice. Narrata offers only the eight preset voices, not custom cloning of individual speakers.
Are the voices based on real people?
Yes. The eight presets are based on licensed recordings from LibriTTS-R (CC BY 4.0), from LibriVox public-domain audiobook readings. These are synthetic voices generated from real human recordings, not entirely computer-generated sounds.
Who owns the rights to the preset voices?
The voices are based on LibriTTS-R (CC BY 4.0), from LibriVox public-domain audiobook readings. Rights depend on the specific source license. Narrata provides access to these presets, but you should verify the underlying license requirements for your specific use case.
Is the text I send to the API used to train the model?
No. Input text is used only to produce the audio and is not used for training. Generated audio is returned in the response and not stored. Only request metadata (time, characters, status) is kept for billing purposes.
Do the generated audio files contain watermarks?
Yes. Every generated file carries an inaudible watermark that marks it as synthetic speech. This helps identify the audio as AI-generated without affecting the listening experience for your audience.
Can I use the audio for commercial purposes?
Yes. The underlying data comes from LibriTTS-R (CC BY 4.0), from LibriVox public-domain audiobook readings. You can use the generated audio commercially, provided you comply with the specific licenses of the source recordings used by the preset voices.
Do I need to attribute the voice actors?
Likely, depending on the source license. Since voices are based on LibriTTS-R (CC BY 4.0) and LibriVox, attribution may be required. Check the specific license terms for the preset voice you select to ensure compliance.
What is LibriTTS-R?
LibriTTS-R is a dataset used for the preset voices. It is licensed under CC BY 4.0, from LibriVox public-domain audiobook readings.
Are the voices public domain?
The source recordings are from LibriVox public-domain audiobook readings. However, the specific dataset LibriTTS-R is licensed under CC BY 4.0. This means you may need to provide attribution depending on how you use the generated audio.
Can I clone my own voice with this API?
No. Custom voice cloning is not offered. You can only choose from the eight preset voices (Aria, Nova, Luna, Sage, Felix, Milo, Atlas, Orion). You cannot upload your own voice to create a unique clone.
Is custom voice cloning offered?
No. The product offers eight preset voices based on licensed recordings. There is no feature to upload a voice sample and create a custom clone. You are limited to the eight available presets for all speech generation.
Understanding Voice Cloning vs. Preset Voices
AI voice cloning typically involves training a model on a specific individual’s unique vocal patterns to replicate their voice. In contrast, our API uses eight fixed preset voices derived from licensed recordings. These presets are static; selecting one does not create a new, unique identity but rather applies a known, standardized vocal profile to your text. Every generated file carries an inaudible watermark that marks it as synthetic speech.
This distinction matters for legal clarity. Because we do not offer custom cloning for customers, your team selects from established voices rather than generating a bespoke replica of a specific person. This simplifies the consent landscape, as the voices are based on pre-licensed source material rather than individual personality rights.
What is Voice Cloning?
Voice cloning creates a unique synthetic voice model trained on specific audio data. This process often requires explicit consent from the voice owner, as the result is a personalized imitation of a distinct individual. Our platform does not provide this custom cloning capability, avoiding the complex rights management associated with replicating specific personalities.
- Custom models trained on unique voice data
- Requires individual consent for specific persons
- Creates a personalized synthetic identity
Preset Voices Are Licensed
Our eight voices are based on licensed recordings from the LibriTTS-R dataset and LibriVox public-domain audiobooks. Each preset is a fixed vocal style, not a clone of a specific celebrity or employee. This approach ensures that the voice source is transparent and legally grounded in established licensing agreements, reducing risk for L&D teams deploying content at scale.
- Based on LibriTTS-R (CC BY 4.0)
- Source: LibriVox public-domain readings
- Fixed preset profiles, not custom clones
Consent Requirements for Synthetic Voices
When cloning a specific person’s voice for commercial use, consent is generally required because the voice itself may be protected under right of publicity laws. Unlike preset voices, which are fixed assets, cloning creates a unique derivative that mimics an individual’s identity, necessitating explicit permission from the voice owner before distribution in your LMS or courses. Narrata does not offer this custom cloning capability.
Preset voices, such as Aria or Felix, do not require individual consent for each use because they are based on licensed recordings from the LibriTTS-R dataset (CC BY 4.0) or public domain LibriVox readings. Every generated file carries an inaudible watermark that marks it as synthetic speech. These voices are generic presets; they do not replicate a specific celebrity or employee, so no additional legal clearance is needed beyond the standard API usage terms.
When Consent Is Needed
If your organization plans to replicate a specific employee’s or narrator’s voice for training materials, you must secure their consent. This is because synthetic speech that mimics a particular individual’s unique vocal characteristics can infringe on their right of publicity, especially in commercial or public-facing educational content.
Consent is critical when the cloned voice is used to represent the individual in a way that implies endorsement or specific association. Unlike generic presets, cloned voices carry a personal identity, making legal clearance necessary before deploying them in large-scale LMS libraries or external course platforms.
- Secure written consent before cloning an employee’s voice for commercial training.
- Clarify usage rights (internal vs. external distribution) in the consent form.
- Retain records of consent for audit trails and compliance verification.
Preset Voices and Licensing
Our preset voices (Aria, Nova, Luna, Sage, Felix, Milo, Atlas, Orion) are based on licensed recordings from the LibriTTS-R dataset, released under CC BY 4.0, or public domain LibriVox readings. These voices are generic and do not represent specific individuals, so no additional consent is required for their use in your educational content.
Since these voices are not tied to a specific person’s identity, they can be used freely within the bounds of the API’s terms. This eliminates the need for individual legal reviews for each voice selection, streamlining the deployment of audio across large course libraries.
- No additional consent needed for preset voices like Aria or Felix.
- Voices are licensed via CC BY 4.0 or public domain sources.
- Generic nature avoids right of publicity concerns common with cloning.
Data Usage and Training
Input text is used strictly to produce the audio and is not used for training. Generated audio is returned in the response and not stored. Request metadata (time, characters, status) is kept for billing.
Input Text Not Used for Training
Your source text for courses is processed to generate the audio file but is not added to our training datasets. This means your specific educational content does not influence future voice models, ensuring your proprietary material remains isolated from general model improvements.
Audio Not Stored
The synthesized audio files are returned directly in the API response and are not retained on our servers after delivery. Only request metadata—such as timestamp, character count, and status—is preserved for accurate billing purposes. This minimizes data exposure for L&D teams managing sensitive curriculum content.
Licensing of Source Recordings
Our synthetic voices are built on licensed recordings from the LibriTTS-R dataset, released under the CC BY 4.0 license. This framework ensures that the foundational audio data used to train our models is legally cleared for commercial use, providing a solid basis for enterprise L&D libraries without unexpected royalty disputes. Input text is used only to produce the audio and is not used for training.
LibriTTS-R and CC BY 4.0
The CC BY 4.0 license allows for broad usage, including commercial applications, provided attribution is given where required by your specific integration. This standard licensing model simplifies compliance for large educational content libraries, as the source material is pre-vetted for structural integrity and voice quality.
- CC BY 4.0 license covers the base audio data
- Suitable for commercial and internal educational use
- Standardized licensing reduces legal overhead for teams
LibriVox Public Domain
Additional voice data originates from LibriVox, which hosts public-domain audiobook readings. These recordings are free of copyright restrictions, further securing the rights for your generated content. This dual-source approach ensures that your LMS audio assets are built on transparent, legally sound foundations.
- LibriVox provides public-domain source material
- No additional licensing fees for these specific voices
- Enhances legal security for distributed course content
Imperceptible Watermarks
Every generated audio file carries an imperceptible digital watermark that marks it as synthetic speech. This technical marker is embedded directly into the audio waveform, allowing downstream systems to identify AI-generated content without altering the listening experience for your learners. The watermark is inaudible to human ears.
Synthetic Speech Marking
The watermark is inaudible to human ears, ensuring that the audio quality remains pristine for your e-learning modules. It serves as a reliable technical signal for content management systems that need to distinguish between human-recorded and AI-synthesized voiceovers.
- Embedded directly into the audio waveform
- Inaudible to the end listener
- Machine-readable by compatible CMS platforms
Transparency in Audio
While the watermark provides technical proof of synthetic origin, it complements your broader transparency strategy. L&D teams can combine this data point with your internal voice selection logs to maintain clear records of which assets are AI-generated.
Best Practices for LMS Teams
LMS teams must maintain clear records of which preset voice was used for each generated audio file. Since Narrata provides eight distinct licensed presets rather than custom clones, tracking the specific voice ID alongside the content ID simplifies compliance audits. This documentation proves that the audio originates from standard, licensed sources like LibriTTS-R, avoiding ambiguity about voice origin.
Internal communication should clarify that these are synthetic voices, not recordings of real employees or celebrities. Since no custom cloning is offered, there is no need to negotiate individual likeness rights. Simply ensuring staff understand that the audio is AI-generated and marked with an inaudible watermark is sufficient for most corporate policies.
Documenting Voice Sources
Maintain a simple log linking each course module to the specific voice preset used (e.g., 'aria' for narrations, 'atlas' for news-style updates). This creates an immediate audit trail for compliance teams.
Verify that the source recordings (LibriTTS-R) and the resulting synthetic output are correctly categorized under CC BY 4.0 or public domain standards, depending on your specific usage rights interpretation.
- Log voice ID per audio asset.
- Store generation timestamps and character counts for billing and audit reference.
- Keep a master list of the 8 available presets and their licensing status.
Employee Communication
Inform content creators that the voices are synthetic and carry inaudible watermarks. This transparency helps prevent internal confusion about whether a real person’s voice was used without permission.
Clarify that since Narrata does not offer custom voice cloning, employees do not need to seek individual consent for likeness rights, simplifying the internal approval process significantly.
- Explain the difference between preset AI voices and human recordings.
- Share the inaudible watermark fact for transparency.
- Clarify that no custom cloning means no likeness rights issues.
Common Misconceptions About Cloning
Teams often assume that any synthetic voice is a "clone" requiring individual consent. In reality, our preset voices are fixed presets derived from licensed recordings, not dynamic replicas of specific individuals. This distinction simplifies compliance for L&D teams, as the voices are pre-cleared assets rather than personalized biometric data.
Cloning vs. Preset Selection
Our eight voices are static presets, not generated from a user’s unique vocal sample. Because they are not custom clones of a specific person, the consent model differs significantly from true voice cloning. You are selecting from a library of licensed assets, not replicating an employee’s or instructor’s exact voice.
- Preset voices are static, not dynamic replicas
- No unique biometric data is replicated
- Simplified consent compared to custom cloning
Custom Voice Options
We do not offer custom voice cloning for customers. There is no option to upload a voice sample to create a unique clone. This absence of cloning capability means you do not need to manage individual consent forms for each voice used in your course library, reducing administrative overhead.

Legal Considerations for Large Teams
Large educational teams must navigate varying regional laws regarding synthetic speech, as consent standards differ by jurisdiction. While our preset voices are licensed, understanding local regulations ensures your L&D content remains compliant when deployed across different markets.
Regional Variations
Legal requirements for synthetic audio vary significantly by region. Teams deploying content globally should verify local consent norms for AI-generated narration, especially in educational contexts where voice likeness matters.
Internal Policy Creation
Establish clear internal guidelines for voice selection and usage rights. Document which preset voices are used for which content types to maintain an audit trail. This ensures consistent legal coverage across large content libraries without requiring custom cloning approvals.
- Define approved preset voices for specific course categories
- Maintain a record of voice selection per module
- Update policies as regional laws evolve
Choosing the Right Voice for Your Audience
Selecting the appropriate voice preset ensures your educational content resonates with learners while maintaining a consistent brand tone. The API provides eight distinct presets, ranging from the warm, clear narration of Aria to the energetic promo style of Felix, allowing teams to match audio to specific course modules or target demographics.
For technical or instructional material, Sage offers a calm, instructional tone ideal for step-by-step guides, while Orion’s deep, documentary style suits deeper dives into complex topics. Matching voice character to content type enhances engagement without requiring custom voice design or additional licensing steps.
Tone and Style Matching
Pairing voice characteristics with content type improves retention. Use Luna for storytelling modules and Milo for conversational explanations. Each voice is engineered for clarity, ensuring that complex educational material remains accessible across all 23 supported languages.
Consistency is key for large LMS libraries. Once a voice is selected for a course series, maintaining that choice helps learners recognize the material instantly. The API’s 4096-character limit per request allows for modular content creation that fits these specific tonal needs.
- Aria: Warm, clear narrator for general courses
- Sage: Calm, instructional for technical guides
- Orion: Deep, documentary for in-depth topics
- Felix: Energetic for promotional or introductory content
Audience Demographics
Consider the age and background of your learners when choosing a voice. Younger audiences may respond better to the bright, upbeat Nova, while professional development modules might benefit from the neutral, news-like Atlas. This selection requires no custom cloning, just preset usage.
All eight voices can speak any of the 23 supported languages, including English, Spanish, German, and Japanese. If the input text uses Latin script but is not English, set the language field explicitly to ensure accurate pronunciation and demographic alignment.
- Nova: Bright, upbeat for younger learners
- Atlas: Neutral, news-style for corporate training
- Mil: Friendly, conversational for soft skills
- All voices support 23 languages
Technical Implementation of Consent
Technical implementation ensures that consent decisions are traceable and verifiable across large educational content libraries. By logging which voice model generated each asset, L&D teams can maintain a clear record of the synthetic voice source for every module.
This data supports internal audits and demonstrates due diligence when distributing content to diverse student or employee groups. It transforms abstract legal concepts into concrete, searchable metadata.
Logging Voice Selection ["Every API request should store the selected voice preset alongside the generated audio file’s metadata. This includes the voice ID (e.g., 'aria', 'milo') and the timestamp of generation. Storing this mapping allows teams to quickly identify which synthetic voice produced specific lesson content.", 'For LMS integration, this metadata can be appended to the audio file’s tags or stored in a central content registry. This ensures that even if files are moved or archived, the voice selection remains associated with the original creation event.'] ['Store voice ID in audio metadata or CMS', 'Record timestamp of synthesis request', 'Map voice preset to content module ID']
Audit Trails ['Maintain a log of all voice selections for regulatory compliance. If a voice’s licensing status changes or a new consent policy is adopted, these trails allow you to identify exactly which assets are affected. This is critical for large teams managing thousands of hours of content.', 'The API’s request metadata (time, characters, status) provides a reliable foundation for these trails. Combining this with your internal voice selection logs creates a complete audit trail from text input to final audio output, ensuring transparency in your synthetic media usage.'] ['Link API request logs to content assets', 'Track voice changes over time', 'Support compliance documentation with timestamps']
Future-Proofing Your Content
As regulatory frameworks evolve, L&D teams must ensure their audio libraries remain compliant without complete re-generation. Since Narrata’s voices are preset and licensed, the consent baseline is established at the voice level rather than per individual voice model, simplifying long-term content governance.
This approach allows teams to scale content production confidently, knowing the underlying voice rights are secured. Future updates to licensing terms for source recordings will be communicated directly, ensuring your audio assets remain valid for enterprise distribution. Balance does not expire.
Scalability of Consent
Unlike custom voice cloning, which requires ongoing consent management for each unique voice profile, preset voices use standardized licensing. This means consent is tied to the voice preset (e.g., Aria, Felix) rather than individual user approvals, reducing administrative overhead as your content library grows from hundreds to millions of characters.
For large teams, this consistency ensures that legal reviews conducted once for a voice preset remain valid across all generated modules. You do not need to re-verify rights for each new course or lesson, provided the voice preset remains unchanged.
- Single consent check per voice preset
- Reduces legal overhead for high-volume content
- Consistent rights management across all modules
Updating Voice Libraries
As the API evolves, new voices or format options may become available. Because audio is generated on-demand via the API, updating your library is a technical configuration change rather than a re-recording process. Teams can switch presets (e.g., from Luna to Sage) to match new brand guidelines or audience preferences without losing historical audit trails.
The API’s versioned endpoints ensure that previously generated audio files remain stable and accessible even if voice characteristics are subtly refined in the background. This stability supports long-term course archives where audio consistency is critical for learner retention.
- API-driven updates ensure audio stability
- Switch presets without re-recording
- Maintain audit trails across voice changes
Contact for Specific Legal Queries
For complex licensing scenarios or specific legal questions regarding synthetic audio deployment, our editorial team provides direct support. Support is by email only. We focus on clarifying how our preset voice licenses and watermarking standards apply to your specific LMS use case, ensuring your compliance documentation is accurate and up-to-date.
Email Support
Reach out via email for detailed inquiries about voice licensing, data retention policies, or specific legal interpretations of synthetic speech usage. Our team responds to each query with specific, fact-based guidance tailored to your team’s scale and regional requirements.
- No phone support available
- Email-only communication channel
- Support for legal and licensing queries
Editorial Guidance
We offer editorial-style explanations of our technology’s legal standing, helping your team understand the distinction between our licensed preset voices and custom cloning models. This guidance ensures your internal policies align with current industry standards for synthetic audio generation.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. This preset is particularly effective for keeping learners engaged in short modules or promotional training content.
23 languages
Any voice speaks any of these languages.
