Is Voice Cloning Different from Preset TTS Voices?

By Dubbing Voice Metrics editorial team · Updated 2026-11-19

This page clarifies the distinction between our preset TTS voices and true voice cloning for localization agencies. We explain why our eight licensed presets—based on LibriTTS-R recordings—offer predictable, legally clear audio without custom training data. You will learn how these fixed voices differ from cloning in cost, speed, and technical requirements.

Readers will discover the licensing implications of synthetic speech, including inaudible watermarks and CC BY 4.0 rights. We outline when agencies might need custom cloning versus when presets suffice, helping teams balance quality, legal compliance, and budget for global dubbing projects.

Is Voice Cloning Different from Preset TTS Voices?

Frequently asked questions

What is voice cloning?

Voice cloning creates a unique synthetic voice by training a model on specific audio samples from a target speaker. This allows for highly customized speech synthesis that mimics the original person’s tone and timbre, distinct from standard preset voices.

Is this API a voice cloning tool?

No. Our API uses eight fixed preset voices based on licensed recordings. We do not offer custom voice cloning, voice uploading, or voice design features. You select from the provided presets rather than training a new voice model.

How do preset voices work?

Preset voices are synthetic models trained on licensed recordings, mapped to specific voice profiles. When you send text via the API, the engine synthesizes speech using these fixed profiles. Each voice has a distinct tone, such as Aria’s warm narration or Felix’s energetic promo style.

Are the voices real people?

Yes, the voices are based on licensed recordings from real people, specifically from the LibriTTS-R dataset sourced from LibriVox. However, the output is synthetic speech generated by our engine. Each generated file contains an inaudible watermark to mark it as synthetic, distinguishing it from a direct recording of the original speaker.

What is the LibriTTS-R dataset?

LibriTTS-R is the dataset containing the licensed recordings used for our preset voices. It is sourced from LibriVox public-domain audiobook readings. This dataset provides the foundational audio samples that define the eight available voice presets in our API.

Do I need consent to use these voices?

The voices are based on LibriTTS-R, which uses LibriVox public-domain audiobook readings. Since the source material is public domain, no additional consent is typically required for usage. However, you should verify specific licensing terms for your commercial context.

Can I clone my own voice?

No. We do not offer custom voice cloning or allow you to upload your own voice. You must use one of the eight preset voices (Aria, Nova, Luna, Sage, Felix, Milo, Atlas, or Orion) provided by the API.

What is an inaudible watermark?

An inaudible watermark is a subtle signal embedded in the generated audio file. It marks the file as synthetic speech, helping to identify it as AI-generated rather than a human recording. This feature aids in transparency and content verification.

How are synthetic voices detected?

Synthetic voices are detected using the inaudible watermark embedded in every generated file. Additionally, the distinct characteristics of the preset voices, based on LibriTTS-R recordings, may be identifiable by specialized analysis tools designed to distinguish synthetic from human speech.

What is CC BY 4.0?

CC BY 4.0 is a Creative Commons Attribution license. It allows the use of the source recordings with proper attribution. Our preset voices are based on recordings licensed under this standard, ensuring legal clarity for users of our TTS API.

Are these voices public domain?

The source recordings from LibriVox are public domain. The dataset is licensed under CC BY 4.0. You can use the voices commercially; attribution requirements depend on your specific usage of the dataset.

Can I use these voices commercially?

Yes, you can use the preset voices commercially. The underlying recordings are from LibriVox public-domain audiobooks. Our API provides these voices for commercial use, with pricing based on character count ($10 per 1 million characters) and no minimum spend.

What is the difference between cloning and presets?

Cloning creates a unique voice model from specific audio samples, offering high customization. Presets are fixed voices (like Aria or Felix) based on existing recordings. We offer only presets, not cloning, meaning you cannot create a custom voice profile for your brand.

Is voice cloning expensive?

Voice cloning typically requires training costs and higher processing power. Since we do not offer cloning, you only pay for preset usage at $10 per 1 million characters. This model is generally more cost-effective than custom cloning solutions.

Do I need to train a model?

No. Our API uses pre-trained preset voices. You simply send text and receive audio. There is no training phase required, unlike voice cloning where you must train a model on your specific audio samples before generating speech.

How long does voice cloning take?

Voice cloning typically requires training on specific audio samples, which involves data collection and processing time. Since we do not offer cloning, you get instant audio generation. Requests are processed sequentially, with typical speed being roughly real-time for long texts.

Can I customize voice parameters?

You can adjust the speed (0.25–4.0) and select the response format (mp3, wav, opus, flac, pcm). You can also optionally set the language. However, you cannot modify timbre, pitch, or emotion tags beyond the preset voice’s inherent characteristics. No SSML or phoneme customization is available.

Is SSML used for voice cloning?

No. We do not support SSML (Speech Synthesis Markup Language). Voice cloning often uses SSML for fine-grained control, but our preset voices rely on simple text input. You cannot use SSML tags to modify pronunciation or pacing in our API.

Defining Voice Cloning in Context

Sound wave illustration for: Defining Voice Cloning in Context

Determining whether synthetic speech requires voice cloning depends on the definition. In localization workflows, agencies often confuse preset models with cloning because both produce non-human audio. Cloning implies training a new model on specific source material, whereas preset voices are static assets.

What Reddit Users Mean by Cloning

Online discussions frequently label any natural-sounding synthetic voice as 'cloned.' This is inaccurate. Reddit users often use the term to describe high-fidelity preset voices that merely sound realistic, rather than voices trained on a unique dataset.

Technical Definition of Voice Cloning

Technically, cloning requires input data to train a distinct voice model. Our API uses eight fixed presets based on licensed recordings. These voices are static; no custom training occurs during synthesis. We do not offer voice cloning or custom voice uploads.

Consent and Licensing Requirements

Preset voices are derived from LibriTTS-R, sourced from LibriVox public-domain audiobooks. This ensures clear licensing without additional consent fees. Custom cloning requires explicit rights to the specific voice being replicated, which varies by jurisdiction and usage.

  • LibriTTS-R dataset origins
  • Public-domain audiobook sources
  • CC BY 4.0 usage rights

Preset Voices: The Current Standard

Sound wave illustration for: Preset Voices: The Current Standard

Our API delivers eight preset voices generated from pre-trained synthetic models. These voices are not clones but distinct, fixed characterizations designed for consistent output across diverse languages. Each model is trained on specific voice profiles, ensuring uniform tone and pacing without variability between generations.

Pre-Trained Synthetic Models

The eight presets—Aria, Nova, Luna, Sage, Felix, Milo, Atlas, and Orion—represent distinct vocal styles. They are built on licensed recordings from the LibriTTS-R dataset. This ensures a stable baseline quality for all text-to-speech requests.

  • Eight distinct voice profiles
  • Trained on LibriTTS-R data
  • Consistent audio quality

Fixed Voice Characteristics

Each voice maintains fixed tonal characteristics regardless of input text. For instance, Felix remains energetic for promos, while Atlas stays neutral for news. Users select one voice per request, and the synthetic output retains those specific traits throughout the audio file.

No Custom Training Required

Agencies do not need to provide training data or wait for model fine-tuning. Voices are immediately available via the /v1/voices endpoint. This allows for instant scaling across multiple localization projects without additional setup time or computational overhead.

  • Instant availability
  • No training data needed
  • Immediate API integration

Key Differences: Cloning vs. Presets

Sound wave illustration for: Key Differences: Cloning vs. Presets

Voice cloning typically requires training on specific audio samples to replicate a unique timbre, whereas our preset voices are pre-trained models with fixed characteristics. Cloning offers high customization but demands input data; presets provide immediate, consistent quality without additional training.

Customization Level

Cloning allows for personalized voice replication, but requires specific source material. Presets like Aria or Felix offer eight distinct, consistent styles ready for immediate use via API.

Training Data Requirements

Custom cloning needs dedicated audio datasets for training. Our preset voices are already trained on licensed datasets like LibriTTS-R, requiring no additional data from the user.

Cost and Time Implications

Cloning involves higher setup costs and time for training. Presets offer instant availability at a flat rate of $10 per million characters, with no minimum spend or monthly fees. A new key starts with a $0.50 trial credit valid for 7 days.

Why Agencies Use Preset Voices

Sound wave illustration for: Why Agencies Use Preset Voices

Agencies require consistent audio output for global localization, where unpredictable voice quality can disrupt viewer immersion. Preset models like Aria or Felix deliver uniform tone across thousands of episodes, ensuring brand coherence without the variance inherent in custom training runs.

Requests are processed one after another, so under load a request may wait in a queue. This stability allows teams to integrate synthesis into automated pipelines with confidence, knowing that delivery times align with standard processing expectations rather than fluctuating based on model complexity.

Predictable Quality

Pre-trained voices like Nova or Atlas maintain consistent tone and pacing. This uniformity prevents the jarring shifts in quality that often occur when multiple custom models are mixed in a single project. Agencies can trust that every character delivers the same vocal profile, regardless of text length or complexity.

The fixed characteristics of these eight presets eliminate the need for extensive quality assurance checks on voice identity. Editors know exactly how a voice will sound before generation begins, reducing revision cycles and allowing faster turnaround for localized content.

  • Consistent tone across all languages and scripts.
  • No variance between generation runs.
  • Uniform pacing and pronunciation standards.

Instant Availability

All eight voices are immediately accessible via the API without waiting for model training or fine-tuning. Teams can begin generating audio for new projects the moment they receive their API key. There is no lead time for voice preparation, enabling rapid prototyping and same-day delivery for urgent localization tasks.

The system supports 23 languages with any voice, allowing agencies to deploy multilingual content instantly. Whether switching from English to Japanese or Arabic to Swahili, the voice characteristics remain recognizable, ensuring brand continuity across markets without technical delays.

  • No training wait time.
  • Immediate access upon key generation.
  • Ready for any of 23 languages.

Lower Operational Cost

With a flat rate of $0.01 per 1,000 characters, agencies can predictably budget for large-scale projects. There are no per-model fees or minimum spend requirements. Failed requests are refunded in full, ensuring that only successful audio generations incur costs. This model scales efficiently from small pilots to enterprise-level dubbing.

The absence of monthly seats or subscription tiers means agencies pay only for what they use. Combined with the ability to split long texts and manage queues effectively, operational overhead remains low. Teams avoid the capital expenditure of maintaining multiple specialized voice models.

  • Flat rate: $0.01 per 1,000 characters.
  • No monthly subscriptions or seat fees.
  • Full refunds for failed requests.

Licensing of Preset Voices

Sound wave illustration for: Licensing of Preset Voices

Our preset voices are built on licensed recordings from the LibriTTS-R dataset, derived from LibriVox public-domain audiobook readings. This foundation ensures that the synthetic voices carry a clear, professional lineage without the legal ambiguity of modern celebrity recordings.

LibriTTS-R Dataset Origins

The underlying audio data comes from LibriVox, a well-known repository of public-domain audiobooks. By leveraging these established recordings, we provide voices that are already vetted for quality and clarity, ensuring consistent performance across our 8 presets.

  • Source: LibriVox public-domain audiobooks
  • Dataset: LibriTTS-R (CC BY 4.0)

CC BY 4.0 Usage Rights

The dataset is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). This allows us to train our models and deliver commercial-grade synthetic speech while maintaining clear attribution standards. Agencies can use the output for global content without worrying about complex royalty structures.

Public Domain Audiobook Sources

Because the source material is in the public domain, we avoid the high licensing fees associated with celebrity or actor voice clones. This keeps the cost predictable at $0.01 per 1,000 characters, making it ideal for high-volume localization projects.

Watermarks and Synthetic Detection

Sound wave illustration for: Watermarks and Synthetic Detection

For localization agencies distributing synthetic voiceovers globally, transparency regarding AI-generated audio is critical. Our API embeds an inaudible watermark directly into every generated audio file, marking the content as synthetic speech without affecting playback quality.

This technology allows platforms to automatically detect whether audio was produced by our TTS engine, ensuring compliance with emerging media regulations and maintaining trust with end-users who expect clear distinctions between human and synthetic narration.

Inaudible Watermarking Technology

Every audio file generated through the API carries a unique, inaudible watermark. This digital signature is embedded during synthesis and remains intact across standard formats like mp3, wav, opus, flac, and pcm.

The watermark is imperceptible to the human ear, preserving the natural listening experience while providing a reliable method for verifying the synthetic origin of the voiceover.

Identifying Synthetic Speech

Third-party detection tools can scan the audio file to confirm the presence of the synthetic watermark. This provides agencies with a verifiable proof of origin, essential for auditing content pipelines.

Unlike visual artifacts, audio watermarks are robust against common compression and format conversions, ensuring detection remains effective across different distribution channels.

Implications for Content Platforms

Major content platforms increasingly require disclosure of AI-generated media. By providing synthetic audio with embedded watermarks, our service simplifies compliance with these transparency mandates.

Agencies can confidently deliver localized content knowing that the synthetic nature of the voices is technically verifiable, reducing legal risk and enhancing consumer trust.

When to Consider Custom Cloning

Sound wave illustration for: When to Consider Custom Cloning

Custom voice cloning is a distinct process from using preset TTS voices. It involves training a model on specific audio samples to replicate a unique timbre, which requires dedicated data collection and processing time. This section outlines scenarios where this additional investment becomes necessary for localization agencies.

Unlike our preset voices like 'atlas' or 'nova', which are fixed models based on licensed datasets, custom cloning creates a proprietary asset. Agencies must weigh the cost of training against the brand value of a unique voice.

Brand Specific Voice Needs

Agencies often require a voice that matches a brand’s exact identity beyond generic presets. Custom cloning allows for the replication of a specific tone, accent, or speech pattern that resonates with a target demographic in a way predefined models cannot.

This is critical for global campaigns where brand consistency is paramount. However, it demands a higher initial investment and ongoing licensing agreements compared to using standard voices like 'sage' or 'milo'.

  • Requires dedicated audio samples for training

Celebrity or Actor Licensing

Using a celebrity or actor’s voice for dubbing requires explicit legal clearance. Unlike our preset voices which are based on LibriTTS-R (CC BY 4.0) or public domain sources, custom cloning of a recognizable voice involves complex rights management.

Agencies must secure permissions from the voice talent or their estate. This process adds legal overhead and potential royalty structures, making it suitable for high-profile productions rather than routine content localization.

  • Requires explicit legal clearance from talent

High-Volume Unique Narration

For projects requiring thousands of hours of narration with a single, consistent voice, custom cloning ensures uniformity. While our preset voices are efficient, they may not match the specific nuance of a lead narrator.

If a project demands a voice that feels personally tailored rather than generically synthetic, cloning offers that uniqueness. However, it is less cost-effective for short-term or variable content compared to using our standard API voices.

  • Ensures voice consistency across large volumes
When to Consider Custom Cloning

Consent and Legal Considerations

Sound wave illustration for: Consent and Legal Considerations

Agencies must verify that synthetic voices derived from licensed datasets carry clear commercial rights for global dubbing. While our preset voices originate from public domain audiobooks, the synthetic output itself is a distinct asset requiring proper licensing for broadcast or streaming.

Understanding ownership ensures that localized content remains free of downstream copyright disputes, especially when voices are used in high-volume commercial campaigns or adaptive media platforms.

Voice Rights Ownership

Synthetic voices are generated from pre-recorded source material, but the resulting audio file is a new creation. Ownership typically rests with the generator or licensee, depending on the platform’s terms. For agencies, this means you hold the rights to the generated MP3/WAV files for distribution, provided you comply with the API’s usage policy.

Unlike human actors, synthetic voices do not require ongoing residual payments or union agreements, simplifying rights management for multi-language releases. However, the underlying source recordings may retain their original copyright status.

  • Generated audio files are owned by the user for distribution.
  • Source recordings may retain original copyright holders.
  • No ongoing residuals for synthetic voice usage.

Public Domain vs. Licensed

Our voices are based on LibriTTS-R (CC BY 4.0) and LibriVox public domain readings. This means the source recordings are free from restrictive copyright claims, allowing broad commercial use without attribution requirements in most jurisdictions.

Public domain status ensures that the base voice characteristics are not encumbered by exclusive licensing deals, making them ideal for global localization where rights clearance can be complex and costly.

  • LibriTTS-R uses CC BY 4.0 license.
  • LibriVox sources are public domain.
  • No exclusive rights restrictions on base voices.

Commercial Usage Rights

All preset voices support commercial use, including advertising, e-learning, and video dubbing. The generated audio can be sold, streamed, or used in paid content without additional per-use fees beyond the API cost.

Since voices are synthetic, they do not infringe on personal publicity rights. This avoids issues with celebrity voice impersonations, provided the voice is not explicitly marketed as a specific celebrity’s likeness.

  • Commercial use allowed for all presets.
  • No per-use fees beyond API cost.
  • No publicity right infringement for synthetic voices.

Future of Voice Cloning in TTS

Sound wave illustration for: Future of Voice Cloning in TTS

The TTS landscape is shifting toward on-demand voice synthesis, reducing the need for weeks of studio recording. However, true high-fidelity cloning remains a specialized, high-cost tier rather than a commodity feature for standard localization pipelines.

While generative models improve, they do not replace the need for explicit consent and licensing, which remains a primary barrier for mass adoption in enterprise dubbing workflows.

Emerging Cloning Technologies

Current technology allows for voice replication with minimal audio input, yet it differs fundamentally from our preset models. Unlike the fixed, licensed presets available via our API, cloning requires dedicated training data and often yields variable consistency across different languages or emotional tones.

Market Trends

The market is seeing a split: general-purpose TTS continues to optimize for cost and speed, while cloning services target premium brand identities. We do not offer custom cloning, so agencies must integrate separate providers if unique voice assets are required.

Agency Adoption Rates

Adoption of custom cloning remains low among mid-sized localization agencies due to the high per-voice cost and legal complexity. Most agencies prefer the predictability of preset voices like Aria or Felix, which offer instant availability and clear licensing terms without additional negotiation.

Summary of Voice Options

Sound wave illustration for: Summary of Voice Options

Preset voices offer immediate deployment with consistent, licensed quality at $0.01 per 1,000 characters, while custom cloning demands specific source material, higher costs, and legal clearance. Agencies must weigh the speed and predictability of the eight available presets against the unique branding value of bespoke audio.

Selecting the right voice strategy depends on volume and brand identity. Presets ensure instant availability and lower operational overhead, whereas cloning provides distinct character for high-value narratives.

Pros of Preset Voices

Presets provide instant availability without training delays. The eight voices cover diverse tones, from warm narrators to energetic promos, suitable for global dubbing. Licensing is clear under CC BY 4.0 and public domain sources, reducing legal overhead.

Cost efficiency is significant at $0.01 per 1,000 characters. With no custom training required, agencies can scale production quickly. The consistent quality across 23 languages ensures uniformity in localized content.

  • Instant deployment
  • Clear licensing
  • Low cost
  • Multi-language support

Pros of Custom Cloning

Cloning allows for unique brand voices and celebrity endorsements. It offers high customization for specific narrative needs, though it requires dedicated training data and legal consent.

This approach is ideal for high-volume unique narration where brand identity is paramount. It distinguishes content from standard synthetic speech, though it incurs higher production costs.

  • Unique brand identity
  • Celebrity licensing
  • High customization
  • Distinct narrative style

Recommendations for Agencies

For most localization projects, presets offer a strong balance of cost, speed, and legal clarity. Reserve cloning for flagship products or specific celebrity endorsements where unique audio adds measurable value.

Always verify licensing for cloned voices. Use presets for bulk content to maintain predictable budgets. Evaluate synthetic detection needs when distributing to platforms requiring transparency.

Common Misconceptions

Sound wave illustration for: Common Misconceptions

Agencies often assume preset voices are merely clones of generic speakers, but they are distinct synthetic models with fixed characteristics. Unlike custom cloning, which requires training data from a specific individual, presets like Aria or Atlas offer consistent, predictable quality without the variability of a cloned voice.

Cloning is not a magic button; it demands significant human input, including source audio, consent, and processing time. Presets, by contrast, are instantly available and require no training phase, making them more efficient for high-volume localization workflows where speed and consistency matter more than unique brand identity.

Preset Voices Are Not Clones

Preset voices are pre-trained models based on licensed datasets, not replicas of specific individuals. They offer a consistent tone without the nuance of a cloned voice, ensuring uniformity across large projects. This distinction is crucial for agencies needing reliable, repeatable results without the variability inherent in custom cloning.

Cloning Requires Human Input

Custom cloning is not instantaneous. It requires source audio, legal consent, and processing time to train the model. Presets, by contrast, are ready immediately. This makes presets more efficient for agencies managing multiple projects simultaneously, where waiting for custom voice training could delay delivery.

  • Requires source audio samples
  • Needs legal consent forms
  • Involves processing time
  • Not instant like presets

Synthetic Speech is Always Detectable

All generated audio files carry an inaudible watermark marking them as synthetic. While detection methods vary, the presence of this watermark is a consistent feature. Agencies should account for this in content platforms where synthetic detection is a requirement, ensuring transparency and compliance with platform guidelines.

Conclusion for Localization Teams

Sound wave illustration for: Conclusion for Localization Teams

For localization teams, preset TTS voices offer a reliable, legally clear, and cost-effective alternative to custom cloning. With eight distinct presets supporting 23 languages, agencies can maintain consistent brand audio without the overhead of training new models or negotiating complex voice rights.

By leveraging the included watermarking and clear licensing under CC BY 4.0, teams ensure compliance while keeping operational costs low. This approach allows for rapid scaling of dubbed content without sacrificing audio quality or incurring the high fees associated with celebrity or custom voice licensing.

Best Practices for Voice Selection

Select voices based on content tone rather than trying to mimic specific human speakers. Use the eight presets (e.g., Atlas for news, Luna for storytelling) to match the genre, ensuring consistent auditory branding across all localized assets.

  • Match voice personality to content genre (e.g., Felix for promos, Sage for instruction).

Balancing Cost and Quality

Preset voices provide high-quality synthetic speech at a predictable rate of $0.01 per 1,000 characters, eliminating the variable costs of custom cloning projects. This model supports high-volume dubbing with instant availability and no minimum spend.

Failed requests are free, ensuring you only pay for successful audio generation. The inaudible watermark also aids in content identification without requiring additional metadata management.

Final Editorial Verdict

For most localization workflows, preset TTS voices are a strong choice. They offer immediate scalability, clear legal standing, and consistent quality without the technical debt of managing custom voice models. Agencies should prioritize these presets for standard content and reserve custom cloning only for unique, high-value brand assets.

Listen to the 8 voices

Each sample below was generated by this API from the same sentence. Split logic should respect word boundaries to avoid cutting mid-sentence.

Avatar of the voice aria
ariafemale
Avatar of the voice nova
novafemale
Avatar of the voice luna
lunafemale
Avatar of the voice sage
sagefemale
Avatar of the voice felix
felixmale
Avatar of the voice milo
milomale
Avatar of the voice atlas
atlasmale
Avatar of the voice orion
orionmale

23 languages

Any voice speaks any of these languages.

Sources and further reading

  1. Wikipedia
  2. IETF RFC 7845: Ogg Opus
  3. Node.js documentation
  4. LibriTTS-R corpus (OpenSLR 141)
  5. FFmpeg documentation
  6. ISO 639 language codes
  7. Python documentation
  8. OpenAI API reference: create speech
  9. W3C WAI: making audio and video accessible
  10. Creative Commons Attribution 4.0
  11. curl documentation
  12. openai-python on GitHub

More on this site