Synthetic Speech API for ML Dataset Generation

By VoxyData editorial team · Updated 2026-11-19

VoxyData helps product teams and ML engineers source high-quality, copyright-clean synthetic audio datasets for voice models. This page explains how we simplify TTS access with a reliable API, transparent character-based pricing, and consistent voice quality for dataset generation.

We built this platform for developers who need scalable synthetic speech without the complexity of licensing negotiations. Read on to understand our voice sources, data privacy practices, and technical infrastructure, or check our pricing and docs to start generating audio for your models.

Synthetic Speech API for ML Dataset Generation

Frequently asked questions

Who runs VoxyData?

VoxyData is an independent storefront for a cloud text-to-speech API. We help product teams and ML engineers source high-quality, copyright-clean synthetic audio datasets. Our API uses transparent character-based pricing and consistent voice quality for dataset generation, operating as a standalone service without ties to larger tech giants.

Are the voices copyrighted?

Our voices are based on licensed recordings from LibriTTS-R, which uses CC BY 4.0 data, and from LibriVox public-domain audiobook readings. This ensures that the synthetic audio you generate for your machine learning models is built on a clean, transparent foundation suitable for commercial and research datasets.

Do you use input data for training?

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. We keep request metadata, such as time, characters, and status, solely for billing purposes, ensuring your data remains private and unused for model training.

How long is metadata kept?

Request metadata, including time, characters, and status, is kept for billing purposes. This allows us to accurately track your usage and charges. The generated audio itself is not stored after being returned in the response, ensuring that your specific input data does not persist on our servers long-term.

Can I contact you by phone?

No, we do not have a phone number. Support is by email only. For any questions regarding your API usage, billing, or technical issues, please reach out via email, and our team will assist you directly through that channel.

Do you have a physical address?

We do not list a physical address. As an independent storefront for a cloud API, we operate digitally. Our focus is on providing reliable synthetic speech datasets for ML engineers. You can find contact information and support details on our /contact/ page for any necessary communications.

Are you affiliated with OpenAI?

No, we are not affiliated with OpenAI. Our API endpoint structure matches the OpenAI speech endpoint, so the official OpenAI SDKs work by changing the base_url. However, we are an independent service with our own pricing, voices, and infrastructure, distinct from OpenAI’s offerings and brand identity.

Do you offer SLA guarantees?

No, we do not offer an SLA or uptime guarantee. We are an independent storefront providing a reliable TTS API for dataset generation. While we aim for consistent performance, we do not have enterprise contracts or formal service level agreements. Our pricing is transparent, with no monthly plans or seats.

Is there a data center location?

We do not specify a data center location. Our cloud API processes requests globally, handling text input and returning audio files. We do not offer on-premise installations or allow customers to choose specific data centers. The service is designed for ease of integration via standard HTTP requests to our API.

Do you have SOC 2 certification?

No, we do not have SOC 2, HIPAA, or GDPR certifications. We are an independent storefront focused on providing synthetic speech datasets. Our data practices are clear: input text is used only for generation, and metadata is kept for billing. We do not hold external compliance certificates.

How do I report a bug?

You can report bugs via email, as we offer support by email only. There is no phone number or live chat. Please include details about your request, such as the model used and the input text, to help us investigate. We do not have a public bug tracker.

Can I request a new voice?

We currently offer eight preset voices based on licensed recordings. We do not offer custom voice cloning, voice design, or uploading your own voice. If you need a specific voice not in our list, you cannot request it directly. Our catalog remains fixed at these eight distinct voice options.

Is the API open source?

No, our API is not open source. We provide a cloud text-to-speech API accessible via standard HTTP requests. While we use a request shape similar to OpenAI, our underlying engine and infrastructure are proprietary. You can use the OpenAI SDK by adjusting the base_url, but the API itself is closed.

Who licenses the LibriVox recordings?

Our voices are based on licensed recordings from LibriTTS-R (CC BY 4.0) and from LibriVox public-domain audiobook readings. These sources provide the foundation for our eight voice presets. We ensure that the synthetic audio generated for your datasets is built on these clean, legally sound audio sources.

Our mission

Sound wave illustration for: Our mission

VoxyData exists to solve a specific bottleneck in synthetic speech production: sourcing consistent, high-quality audio for machine learning datasets. We provide a reliable TTS API that allows product teams and ML engineers to generate labeled speech data.

Our goal is to simplify the TTS access pipeline. By providing a straightforward API with transparent character-based pricing, we remove the friction of negotiating custom licenses or managing complex infrastructure. You get consistent voice quality across thousands of files, ensuring your training data remains uniform and your models perform reliably.

Supporting ML datasets

Supporting ML datasets

We focus on generating the clean, labeled audio files that modern ML models require. Our API is optimized for volume and consistency, allowing you to create diverse training sets quickly. We provide the raw material your algorithms need to learn effectively.

By standardizing the input-output loop, we help teams avoid the variability that often plagues scraped audio datasets. Every file is generated through the same engine, ensuring uniform audio quality characteristics across your entire dataset.

  • Generate thousands of labeled audio files on demand
  • Ensure uniform audio quality across your entire dataset
  • Optimized for volume and consistency in ML pipelines

Simplifying TTS access

Simplifying TTS access

Accessing high-quality synthetic speech shouldn't require a dedicated infrastructure team. We offer a simple REST API that fits directly into your existing workflows. With standard JSON requests and OpenAI SDK compatibility, you can integrate TTS generation into your data processing pipelines with minimal code changes.

We handle the complexity of voice synthesis so you can focus on your model. Our API accepts text and returns audio files directly. Just send the text, receive the audio, and feed it into your training loop.

  • REST API with standard JSON requests
  • Compatible with OpenAI SDKs for easy integration
  • Handles synthesis complexity for you

Transparent technology

Transparent technology

We believe in clear pricing and predictable costs. Our character-based model means you only pay for what you generate, with no hidden fees for processing time or file formats. This transparency allows you to budget accurately for your dataset generation projects without worrying about unexpected charges.

Our technology is straightforward and open. We use recordings from LibriTTS-R under a CC BY 4.0 license, ensuring that your synthetic audio is copyright-clean. You know exactly where the voice data comes from and how it is licensed, making it safe for commercial use in your ML products.

  • Character-based pricing with no hidden fees
  • Licensed recordings from LibriTTS-R (CC BY 4.0)
  • Clear licensing for commercial use in ML products

Voice licensing

Sound wave illustration for: Voice licensing

VoxyData’s synthetic voices are built on licensed recordings from the LibriTTS-R dataset, ensuring a copyright-clean foundation for your ML models. This approach provides consistent voice quality without the legal ambiguity often found in scraped audio data.

LibriTTS-R source

LibriTTS-R source

The eight voice presets—Aria, Nova, Luna, Sage, Felix, Milo, Atlas, and Orion—are derived directly from the LibriTTS-R corpus. This dataset offers a variety of tonal characteristics, from warm narrators to energetic promos.

  • Eight distinct voice presets
  • Diverse tonal ranges

CC BY 4.0 license

LibriTTS-R is licensed under Creative Commons Attribution 4.0 (CC BY 4.0). This allows VoxyData to distribute the synthetic voices legally, providing product teams with clear usage rights for generating training data without additional licensing friction.

Public domain recordings

The underlying recordings originate from LibriVox, comprising public-domain audiobook readings. This ensures that the raw audio material is free from restrictive copyright claims, allowing VoxyData to offer reliable, synthetic audio for commercial and experimental ML datasets.

Data privacy

Sound wave illustration for: Data privacy

We prioritize clarity on how your data moves through our system. When you send text to our API, the input is processed solely to generate the requested audio file. Once the synthesis is complete, the generated audio is returned directly in the response and is not stored on our servers. This ensures that your specific dataset content remains transient and secure during generation.

Our architecture is designed for efficiency without compromising on data integrity. We do not retain the input text after the transaction is finished, keeping your synthetic speech assets under your control. This straightforward approach minimizes exposure and aligns with typical ML dataset workflows where volume and speed are key.

Input text usage

Your input text is used exclusively for the immediate purpose of synthesis. It is not archived or repurposed for any other function within our system. The text is processed to create the audio waveform and is then discarded, ensuring that your unique dataset content is not held longer than necessary.

This single-purpose usage model means you do not need to worry about your proprietary phrases or training data lingering in our storage systems. The audio is generated and delivered to you, completing the data lifecycle for that specific request.

  • Text is used only to produce the audio
  • Input text is not stored after generation
  • No secondary use of input data

No training on input

We do not use your input text or the generated audio to train our base models. The voices you use are based on licensed recordings from LibriTTS-R, and the synthesis engine remains static relative to your specific data inputs. This guarantees consistency in voice quality across your dataset generation projects.

Since your data does not influence the model’s training, the voices you generate will always behave consistently. This is critical for ML teams who need uniform synthetic speech characteristics across thousands of samples without worrying about model drift or data leakage affecting future generations.

  • Input text is not used for model training
  • Generated audio is not used for training
  • Model behavior remains consistent regardless of your data

Metadata retention

While the content of your speech is not stored, we do retain request metadata for billing and operational purposes. This includes the timestamp, character count, voice used, and request status. This minimal data allows us to accurately calculate your usage-based charges and ensure reliable service delivery.

This metadata is kept solely for accounting and troubleshooting. It does not contain the actual text or audio content, providing a transparent record of your API usage without exposing the details of your synthetic dataset.

  • Timestamps, character counts, and status are retained
  • Used for billing and operational reliability
  • Does not include content of text or audio

Technical infrastructure

Sound wave illustration for: Technical infrastructure

VoxyData operates as a cloud-native text-to-speech API, designed for reliability and ease of integration. We provide a straightforward REST API that handles the heavy lifting of audio synthesis, allowing product teams to focus on dataset creation rather than infrastructure management.

The service is accessible via standard HTTP methods, ensuring compatibility with existing development pipelines and tools used by ML engineers.

Cloud API

Our API is hosted in the cloud, providing consistent access from anywhere. You send a POST request with your text and parameters, and receive an audio file in return. This model eliminates the need for local hardware dependencies and scales with your dataset generation needs.

REST endpoints

We offer dedicated endpoints for managing your synthetic speech resources. The primary endpoint handles audio generation, while separate endpoints allow you to list available voices, check model details, and verify language support. This structure simplifies debugging and integration testing.

Standard protocols

The API uses standard JSON for request bodies and supports common audio formats like MP3, WAV, OPUS, FLAC, and PCM. Authentication is handled via standard Bearer tokens, making it easy to integrate with existing API gateways and CI/CD pipelines.

Technical infrastructure

Editorial team

Sound wave illustration for: Editorial team

Our editorial team curates voice presets and validates output quality for machine learning pipelines.

VoxyData editorial team

Our editorial team curates voice presets and validates output quality for machine learning pipelines.

Technical writers

We document the API specifications, rate limits, and character counting rules to ensure developers can integrate VoxyData into their data generation workflows without guesswork.

Product insights

We share updates on supported languages and voice availability, helping product teams plan dataset generation cycles around our 23 supported languages and 8 voice presets.

Contact and support

Sound wave illustration for: Contact and support

We keep support simple and direct. For any questions about the API, pricing, or dataset generation workflows, our team is available to help via email.

Email support

Send your inquiries to our dedicated support address. We handle all technical and billing questions through this channel.

  • Direct email access
  • No ticket queues

No phone number

We do not maintain a phone line. This allows us to focus resources on building a reliable API and maintaining transparent pricing for ML teams.

Response time

We aim to reply to all emails promptly. While we do not offer an SLA, you can expect a timely response during business hours.

Listen to the 8 voices

Each sample below was generated by this API from the same sentence. This helps prevent overfitting and improves model generalization.

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. LibriTTS-R corpus (OpenSLR 141)
  3. Python documentation
  4. openai-node on GitHub
  5. TRON developer hub: TRC-20
  6. openai-python on GitHub
  7. LibriVox public-domain audiobooks
  8. FFmpeg documentation
  9. IETF RFC 7845: Ogg Opus
  10. Tether (USDT)
  11. ISO 639 language codes
  12. Node.js documentation

More on this site