Frequently asked questions
Who runs SynthEval?
SynthEval is an independent storefront for a cloud text-to-speech API. Accounts are created per site. We are not affiliated with OpenAI, ElevenLabs, Google, Amazon, Microsoft, or any other brand mentioned in comparisons.
Where do the voices come from?
Our eight voice presets are based on licensed recordings from LibriTTS-R (CC BY 4.0) and public-domain audiobook readings from LibriVox. These sources provide the foundational audio used to generate the synthetic speech output delivered via our API endpoints.
Are the voices CC BY 4.0 licensed?
The underlying recordings are from LibriTTS-R (CC BY 4.0) and LibriVox public-domain readings. However, every generated audio file carries an inaudible watermark marking it as synthetic speech. We do not offer custom voice cloning or voice design services.
Do you store my input data?
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. We keep request metadata, such as time, characters, and status, strictly for billing purposes.
How can I contact support?
Support is available by email only. We do not have a phone number, live chat, or on-site support team. You can reach us via the contact page or by emailing the support address provided in your account dashboard.
Do you have a physical office?
We do not list a physical office location. As an independent API provider, we operate remotely. There is no on-premise install option, and we do not offer enterprise contracts or dedicated data-center choices for our synthetic speech services.
Are you affiliated with OpenAI?
No. We are not affiliated with OpenAI, ElevenLabs, Google, Amazon, Microsoft, or any other brand. Our API endpoint shape mimics OpenAI’s speech endpoint for compatibility, but we use our own keys, infrastructure, and voice models for synthetic speech generation.
Do you offer voice cloning?
No. We do not offer custom voice cloning, uploading your own voice, or voice design. We provide eight preset voices based on licensed recordings. Each generated file includes an inaudible watermark to identify it as synthetic speech.
What is your data retention policy?
Input text is not used for training and is not stored after generation. We retain request metadata (time, characters, status) solely for billing and usage tracking. This data supports your account balance and character count calculations.
Do you have an SLA?
No. We do not provide an SLA, uptime guarantee, or enterprise contract. Requests are processed sequentially, so under load, a request may wait in a queue.
Are you certified (SOC 2, HIPAA)?
No. We do not hold certifications such as SOC 2, HIPAA, or GDPR certificates. We are an independent provider without enterprise-grade compliance documentation. Our focus is on delivering high-quality synthetic speech datasets for machine learning training pipelines.
Do you support mobile SDKs?
No. We do not offer a mobile SDK, a WordPress plugin, or a browser extension. You can use the official OpenAI SDKs by changing the base URL, or use plain HTTP requests to interact with our API.
Is there a WordPress plugin?
No. We do not offer a WordPress plugin, mobile SDK, or browser extension. Integration is done via our REST API using standard HTTP requests or the OpenAI SDKs by adjusting the base URL to our API endpoint.
How do I verify the license terms?
Our voices are based on LibriTTS-R (CC BY 4.0) and LibriVox public-domain readings. Each generated file includes an inaudible watermark marking it as synthetic speech. You can review these details in our documentation or contact us via email for specifics.
Our Mission
SynthEval provides high-quality, watermark-marked synthetic speech APIs designed specifically to fuel machine learning training pipelines. We help startups generate labeled audio datasets at scale without the licensing overhead of traditional voice talent, ensuring every generated file carries an inaudible watermark to mark it as synthetic speech.
Enabling scalable synthetic audio for ML
Our API allows you to send text and receive audio files via a standard cloud endpoint. With support for 23 languages and 8 distinct voice presets, you can generate labeled audio data at scale. Requests are processed one after another, handling up to 4096 characters per call, which simplifies batch processing for large-scale model training.
- 60 requests per minute per key
- 8 voice presets (4 female, 4 male)
- 23 supported languages
- Standardized JSON request format
Simplifying dataset creation with APIs
We replace complex licensing negotiations with a simple API call. By using the same request shape as the OpenAI speech endpoint, our tool integrates directly into your existing infrastructure. You can control voice, speed, and response format via JSON, allowing for automated, programmatic dataset generation without manual intervention.
- OpenAI SDK compatibility
- JSON body requests
- Multiple audio formats (mp3, wav, opus, flac, pcm)
- Speed control from 0.25x to 4.0x
Supporting startups in AI development
Startups can generate labeled audio datasets at scale without licensing overhead. Our pricing model charges $10 per million characters, with failed requests refunded in full. This transparency helps you budget accurately for your AI development cycle, ensuring you pay only for successful synthesis.
- $0.01 per 1,000 characters
- Failed requests are free
- No minimum spend or monthly plans
- USDT (TRC-20) top-ups only
Voice Licensing and Sources
We synthesize audio using eight voice presets derived from the LibriTTS-R dataset, ensuring high-quality synthetic speech for ML training pipelines. Our voices cover 23 languages and support standard TTS parameters like speed and format selection.
Voices from LibriTTS-R dataset
Our eight voice presets—such as Aria, Nova, and Atlas—are based on licensed recordings from the LibriTTS-R dataset and LibriVox public-domain audiobook readings. This source provides diverse acoustic characteristics suitable for generating labeled audio datasets at scale.
- Based on LibriTTS-R (CC BY 4.0)
CC BY 4.0 license terms
The underlying voice recordings are licensed under Creative Commons Attribution 4.0 International. This allows startups to generate commercial-grade audio datasets without complex licensing overhead, provided proper attribution is maintained in your documentation.
Recordings from LibriVox public domain
The source material originates from LibriVox public domain audiobook readings. Each generated file carries an inaudible watermark marking it as synthetic speech, aiding in dataset transparency and model evaluation for machine learning applications.

Data Policy
SynthEval processes text exclusively to generate audio files for your ML pipelines. Once the audio is returned in the response, the input text is not stored or retained on our servers.
Input text used only for generation
Your input text is processed in real-time to produce the audio file. We do not archive the textual content after the synthesis is complete, ensuring your source data remains ephemeral.
No training on user input data
We do not use your submitted text to train our voice models. The synthetic speech generated is based on our existing licensed recordings, and your specific dataset inputs remain separate from our training corpus.
Metadata retained for billing only
We retain request metadata, including timestamps, character counts, and status codes, solely to calculate usage and process charges accurately. This data is kept only as long as necessary for billing transparency and dispute resolution.
Technical Infrastructure
SynthEval operates as a cloud-based REST API, designed specifically for integrating synthetic speech into automated ML pipelines. The service processes requests through a standardized endpoint, ensuring consistent performance for dataset generation without requiring custom infrastructure management on the client side.
Cloud-based API architecture
Our architecture delivers audio files via simple HTTP responses, allowing seamless integration with existing data collection workflows. The system handles character limits up to 4096 per request, requiring clients to split longer texts for large-scale dataset creation.
Standard rate limits and queueing
Requests are processed sequentially to maintain audio quality. Under load, requests may enter a queue, with typical generation speed matching real-time (approx. 25 characters per second). The standard rate limit is 60 requests per minute per key, ensuring stable delivery for continuous training pipelines.
Failed requests incur no charge, and the system returns the audio file directly in the response.
Support for OpenAI SDK compatibility
The API endpoint mirrors the OpenAI speech endpoint structure. By updating the base_url to https://api.play98f.com/v1 and using your SynthEval key, you can utilize official OpenAI SDKs directly. This reduces integration time for teams already familiar with OpenAI’s request shapes and error handling.
- Compatible with OpenAI Python and Node.js SDKs
- Standard JSON request body structure
- Instant key rotation support via the account page

Editorial Standards
We prioritize precision over promotion. Our reviews cite specific technical constraints, such as the 4096-character input limit per request and the 60 requests per minute rate cap, rather than using vague superlatives. Every claim is grounded in the actual API behavior, ensuring that engineering teams can trust the documentation for integration planning.
Honest, specific reviews without hype
We avoid subjective terms like "best" or "most realistic." Instead, we detail concrete metrics: generation speed at roughly 25 characters per second, or the exact $0.01 cost per 1,000 characters. This specificity helps ML engineers estimate inference costs and latency accurately without marketing distortion.
Fact-based comparisons with dates
When comparing SynthEval to other TTS providers, we reference verifiable data points from specific dates. For instance, we note the current voice count (8 presets) and supported languages (23) as of the review date. This ensures that readers can verify the scope of the API against its current state.
No imitation of celebrity voices
Our voices are derived from the LibriTTS-R dataset and LibriVox public-domain recordings. We do not offer celebrity voice cloning or imitation. The eight presets (e.g., Aria, Felix, Atlas) are distinct synthetic voices, ensuring clear distinction between licensed synthetic speech and impersonated human voices.

Contact and Support
SynthEval relies on asynchronous email communication for all customer inquiries, ensuring that technical and billing questions are handled with care and documentation. We do not operate a live chat widget, allowing our team to focus on deep, thoughtful responses rather than quick, shallow interactions.
Email support only, no live chat
We prioritize quality over speed for complex technical queries. Our email support handles API integration issues, billing discrepancies, and dataset generation questions. This approach ensures that every ticket is fully researched before a response is sent, providing you with accurate, actionable information without the noise of real-time chat interruptions.
No phone number available
There is no dedicated phone line for SynthEval support. This decision aligns with our remote-first, startup-oriented model, keeping overhead low and passing those savings on to you. You can reach us at [email protected] for any matter requiring human attention, from voice selection advice to rate limit clarifications.
Response times for inquiries
We aim to respond to support emails promptly. Due to the asynchronous nature of our communication, you will receive a detailed reply rather than a hold-time estimate. For urgent billing issues regarding USDT transfers, please include your transaction hash in the subject line for faster tracking.
- Business day response time: within 24 hours
- Urgent billing: include USDT transaction hash in subject
- No 24/7 live agent availability
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. Our synthetic speech datasets help startups train smarter AI models without expensive voice talent.
23 languages
Any voice speaks any of these languages.
