Frequently asked questions
Who runs the Polyglot API website?
Polyglot API is an independent storefront for a single text-to-speech API. We operate this specific site at play15q.com to help indie developers find the right engine. We are not affiliated with OpenAI, ElevenLabs, or any other brand. Our editorial team focuses on technical metrics like speed, cost, and voice quality for language-learning platforms.
How are the voices licensed?
Our eight voice presets are based on licensed recordings from LibriTTS-R, which is released under the CC BY 4.0 license. These recordings originate from LibriVox public-domain audiobook readings. This licensing ensures that the synthetic speech generated by our API is derived from legally sourced, high-quality source material suitable for commercial and educational use.
Do you use user text for training AI?
No. Input text is used exclusively to produce the requested audio file. We do not use your text data for training our models or improving the engine. This ensures that your specific content remains private and is not repurposed for algorithmic improvement, keeping your educational or commercial materials distinct from our training datasets.
Do you store the generated audio files?
No. Generated audio is returned directly in the API response and is not stored on our servers. We retain only request metadata, such as timestamp, character count, and status, specifically for billing purposes. This minimal data retention ensures that your generated speech files are transient and not archived for long-term storage or analysis.
How can I contact support?
Support is available via email only. We do not offer a phone number, live chat, or on-site contact form for general inquiries. For technical issues or billing questions, please reach out through the email address provided on our contact page. We aim to respond to all support tickets promptly to assist indie developers.
Do you have an SLA?
We do not offer a Service Level Agreement (SLA) or uptime guarantee. Our service is provided on a best-effort basis without enterprise contracts or on-premise installation options. Requests are processed sequentially, so under heavy load, synthesis may experience queue delays. This model allows us to maintain competitive pricing without the overhead of guaranteed uptime tiers.
Are the voices cloned from real people?
No, we do not offer custom voice cloning for customers. You cannot upload your own voice or design a new one. The eight available voices are presets based on existing licensed recordings from LibriVox. While they sound realistic, they are not direct clones of specific celebrities or politicians, and no cloning feature is currently available.
What is the CC BY 4.0 license?
CC BY 4.0 is a Creative Commons Attribution license that allows others to distribute, remix, adapt, and build upon the work, even commercially, as long as they credit the original author. Our voices are sourced from LibriTTS-R under this license, ensuring legal clarity for users who integrate our synthetic speech into their language-learning applications and products.
Who We Are
We are an indie developer-focused editorial team built to cut through TTS noise. We test real-world metrics like speed, cost, and voice quality specifically for multi-lingual education platforms.
Indie developer-focused editorial team
We prioritize technical details that matter for language-learning apps, ensuring your integration works seamlessly without hidden constraints.
Focus on language learning industry
Our benchmarks reflect the unique demands of educational software, such as handling diverse scripts and maintaining consistent pronunciation across 23 languages.
Expert reviews and benchmarks
We publish transparent reviews based on actual API performance. Our data helps you choose the right engine by comparing real character costs, generation latency, and voice clarity.
Voice Licensing and Sources
Our voice engine relies on licensed recordings sourced from public-domain audiobook readings available on LibriVox. By using these established recordings, we ensure that the synthetic voices carry natural prosody and authentic human cadence, which is critical for language-learning applications where clear, realistic pronunciation aids student acquisition.
LibriTTS-R license (CC BY 4.0)
The underlying dataset for our voices is the LibriTTS-R dataset, released under the Creative Commons Attribution 4.0 International license. This allows us to legally process and synthesize the audio data while maintaining transparency about our data sources. The CC BY 4.0 license requires appropriate attribution, which we provide in our documentation.
- Dataset released under CC BY 4.0
- Allows commercial use with attribution
- Standard academic and industrial license
LibriVox public-domain readings
We source our base recordings from LibriVox, a platform hosting public-domain audiobooks. These readings are performed by volunteers and are free of copyright restrictions in the United States and many other jurisdictions. This public-domain foundation ensures that our voice presets are built on stable, legally clear audio material.
- Volunteer-performed readings
- Public-domain audio source
- Stable legal foundation
Licensed recordings usage
Each of our eight voice presets is derived from these licensed recordings. We process the audio to generate our TTS models, ensuring that the output retains the character of the original readings. The generated audio files include an inaudible watermark to mark them as synthetic speech, providing an additional layer of provenance for your users.
- Eight distinct voice presets
- Derived from licensed source material
- Includes inaudible synthetic watermark
Data Policy and Privacy
Polyglot API is designed for developers who value data privacy in educational contexts. When you send text for synthesis, we process it solely to generate the audio response. This input text is never used to train our models or improve the engine, ensuring your content remains private and secure.
Once the audio file is generated and returned in the response, it is not stored on our servers. We retain only the request metadata—such as timestamp, character count, and status—for billing purposes. This minimal data retention ensures accurate usage tracking without storing the actual audio content or input text long-term.
Input text not used for training
Our TTS engine processes your input text exclusively to produce the requested audio file. Unlike many modern AI services, we do not use your submitted text to train or fine-tune our underlying models. This means your unique educational content, whether it’s a language lesson or a story, remains yours and does not become part of our training dataset.
Audio not stored after generation
The generated audio file is returned directly in the API response and is not persisted on our servers. Once your application receives the audio data, we do not keep a copy of the generated speech. This ephemeral storage model ensures that your users’ audio content is not archived in our infrastructure, reducing privacy risks for language-learning apps.
Metadata kept for billing only
We retain minimal request metadata, including the time of the request, the number of characters processed, and the status of the synthesis. This data is used strictly for billing accuracy and to help you monitor your usage patterns. No other personal or content-related data is stored beyond what is necessary to process your request and invoice you correctly.
- Timestamp of request
- Character count processed
- Request status (success/fail)
- No input text storage
- No audio storage
Honest Limitations
The API returns complete audio files upon request completion. We do not support WebSocket connections or partial audio streaming, so you must handle buffering and concatenation if you require continuous playback in your application.
No streaming or real-time audio
The API returns complete audio files upon request completion. We do not support WebSocket connections or partial audio streaming, so you must handle buffering and concatenation if you require continuous playback in your application.
No custom voice cloning
We offer eight preset voices derived from licensed recordings, but we do not provide custom voice cloning or the ability to upload your own audio samples to create unique voice profiles.
No SLA or uptime guarantee
As an independent service, we do not offer formal Service Level Agreements or guaranteed uptime percentages. Our infrastructure is robust, but we do not contractually promise specific availability metrics for enterprise or production use.

Contact and Support
We handle all support requests via email. This asynchronous approach allows us to provide detailed, thoughtful responses tailored to your technical needs, particularly regarding voice selection and integration quirks specific to educational apps.
Email support only
All support requests are handled via email. This allows our team to provide detailed, thoughtful responses tailored to your technical needs, particularly regarding voice selection and integration quirks specific to educational apps.
No phone number
We do not maintain a phone line. As an indie-focused service, we find that asynchronous email communication is more efficient for developers who may be in different time zones or deep in coding workflows.
No live chat
There is no live chat widget on our site. This avoids the common pitfall of quick, shallow answers in favor of comprehensive solutions. You can expect a direct response from a human who understands the API’s nuances.
Our Mission
We exist to help indie developers build better language-learning apps. Our editorial team rigorously tests the API against common workflows, comparing engine performance and output quality so you can integrate a reliable TTS solution without guessing.
Help developers choose the right API
We evaluate latency and generation speeds to ensure your app feels responsive. By analyzing real-world metrics, we help you select the right engine for your specific educational use case.
Transparent pricing and metrics
We believe you should know exactly what you are paying for. Our pricing is straightforward: $10 per 1 million characters. We provide clear metrics on character usage and generation times, ensuring no hidden costs or complex tier structures.
Focus on education use cases
Language learning requires specific audio characteristics: clarity, accurate pronunciation across 23 languages, and consistent voice quality. We prioritize these needs in our reviews, helping you build apps that sound natural and support diverse learners effectively.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. Language learning empowers indie developers to connect deeply with global audiences without heavy marketing budgets.
23 languages
Any voice speaks any of these languages.
Sources and further reading
- Wikipedia
- FFmpeg documentation
- TRON developer hub: TRC-20
- IETF RFC 7845: Ogg Opus
- Creative Commons Attribution 4.0
- curl documentation
- Python documentation
- LibriTTS-R paper (arXiv 2305.18802)
- LibriVox public-domain audiobooks
- openai-python on GitHub
- W3C WAI: making audio and video accessible
- MDN: Web Speech API
