Frequently asked questions
Is this a replacement for Google TTS?
We are not affiliated with Google. Our API provides a reliable text-to-speech endpoint for indie developers building ML datasets. Instead of scraping or manual recording, you get structured audio files instantly. It works with standard HTTP requests and OpenAI SDKs, offering consistent, watermark-marked synthetic voices for your machine learning projects.
What is synthetic speech data?
Synthetic speech data is audio generated by algorithms rather than human recording. Our API produces these files on demand. Each file carries an inaudible watermark marking it as synthetic. This allows you to generate large, consistent datasets for training ML models without the cost and variability of human voice actors or manual recording.
How is the data licensed?
The voices are based on licensed recordings from LibriTTS-R (CC BY 4.0) and LibriVox public-domain audiobook readings. The generated audio files are returned in the response and not stored by us. Input text is used only for synthesis and not for training. You receive the audio file directly for your use.
Can I use the audio for commercial ML models?
Yes. The API provides structured audio files instantly. Since the voices are based on CC BY 4.0 and public-domain sources, the synthetic output is suitable for commercial use. The audio files are returned directly in the response, allowing you to integrate them into your commercial machine learning datasets and models immediately.
What languages are supported?
We support 23 languages: Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, and Turkish. Any voice can speak any of these languages. If the language is not set, it is guessed from the script.
Does the API support streaming?
No, we do not offer streaming, WebSocket, or real-time partial audio. Requests are processed one after another. You send a POST request with text and receive the complete audio file in the response. This synchronous approach is ideal for batch processing and building consistent datasets rather than real-time playback applications.
What is the character limit per request?
Each request supports up to 4096 characters of input text. For longer texts, you should split the content into parts of this size and join the resulting audio files. Spaces and punctuation count toward the character limit. This ensures consistent processing and allows you to manage large dataset generation efficiently.
How do I get API access?
Sign up on our /get-api-key/ page using Google or email. Accounts are per site. The page displays your API key, balance, and a ready curl example. You can start immediately. New keys can be issued anytime, stopping the old one instantly. Support is available via email only.
What formats are available for download?
You can request audio in mp3, wav, opus, flac, or pcm formats. Specify the desired format in the response_format field of your JSON body. The audio is returned directly in the API response. This flexibility allows you to choose the best format for your specific machine learning training pipelines.
Is there a rate limit?
Yes, there is a limit of 60 requests per minute per key. Before your first top-up, the limit is 10 requests per minute. Requests are processed sequentially, so under load, a request may wait in a queue. This ensures stable generation for your dataset creation workflows without overwhelming the infrastructure.
Do you offer voice cloning?
No, we do not offer custom voice cloning, voice design, or uploading your own voice. We provide 8 preset voices based on licensed recordings. These include female voices like Aria and Nova, and male voices like Felix and Milo. You select from these presets to generate your synthetic speech data.
How are voices selected?
You choose from 8 presets: Aria, Nova, Luna, Sage, Felix, Milo, Atlas, and Orion. Each has a distinct tone, such as warm narrator or deep documentary. Select the voice field in your request body. All voices can speak any of our 23 supported languages, allowing flexible dataset generation.
What happens if a request fails?
If synthesis fails, the charge is returned in full. Failed requests are free. This ensures you only pay for successful audio generation. The API returns an error status, allowing you to retry or adjust parameters. This reliability is crucial for automated dataset pipelines where consistent billing and output are required.
Can I control the speaking speed?
Yes, you can adjust the speaking speed between 0.25 and 4.0. Set the speed field in your JSON body. This allows you to generate audio at different rates for your ML models. Faster speeds reduce generation time, while slower speeds may improve clarity for certain training datasets.
First request
curl
curl https://api.play09bet.com/v1/audio/speech \
-H "Authorization: Bearer $TTS_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"tts-1","voice":"aria","input":"Hello from the API.","response_format":"mp3"}' \
--output hello.mp3Why Synthetic TTS Data?
Building high-quality ML datasets often means wrestling with inconsistent human recordings. Synthetic TTS data solves this by providing clean, uniformly generated audio that eliminates the label noise inherent in scraped or manually recorded speech.
Eliminate Label Noise
Human recordings vary wildly in background noise, microphone quality, and speech patterns. Synthetic voices deliver pristine, consistent audio every time, ensuring your training data is clean and your models learn the right features without distraction.
- Uniform background noise profiles
- Consistent microphone simulation
- No human speech artifacts
Scale Data Collection
Generate thousands of hours of speech in minutes. Our API processes requests sequentially but efficiently, allowing you to build massive datasets without hiring voice actors or managing recording studios. Scale your data collection instantly.
- Instant generation of large volumes
- No human scheduling constraints
- Rapid iteration of dataset variants
Consistent Voice Quality
Maintain identical voice characteristics across your entire dataset. Whether you need a warm narrator or a deep documentary voice, every file matches the same quality standard, crucial for robust model training and evaluation.
- Identical voice characteristics
- Uniform audio quality
- Predictable generation times
Voice Presets for Diverse Data
For synthetic speech datasets, consistent voice characteristics are critical for reducing label noise in ML training. Our eight licensed voice presets provide structured, high-quality audio samples that maintain uniform timbre and pacing across large volumes of generated text.
Female Narrators (Aria, Nova)
Aria offers a warm, clear narration style ideal for general-purpose text, while Nova provides a bright, upbeat tone suited for dynamic content. Both voices ensure consistent audio profiles for training models that require distinct female vocal characteristics.
- Aria: Warm, clear narrator for general text
- Nova: Bright, upbeat tone for dynamic content
Male Conversationalists (Milo, Felix)
Milo delivers a friendly, conversational style perfect for interactive datasets, whereas Felix brings an energetic promo style for high-intensity samples. These presets help isolate specific vocal behaviors in training data for conversational AI models.
- Milo: Friendly, conversational style for interaction data
- Felix: Energetic promo style for high-intensity samples
Documentary & News Styles (Atlas, Orion)
Atlas provides a neutral, news-ready delivery, while Orion offers a deep, documentary tone. These voices are optimized for formal speech patterns, making them essential for training ASR systems on structured, professional audio content.
- Atlas: Neutral, news-ready delivery
- Orion: Deep, documentary tone for formal content

Technical Specs for Developers
We provide a reliable, structured API endpoint for indie developers building ML datasets. By sending text via JSON, you receive consistent, watermark-marked synthetic voices instantly, eliminating the need for manual recording or scraping.
OpenAI SDK Compatible
Our endpoint mimics the OpenAI speech structure, so you can use the official OpenAI SDKs by simply changing the base_url to https://api.play09bet.com/v1 and swapping in your key. This allows seamless integration into your existing Python or Node.js pipelines without rewriting request logic.
- Standard JSON request body with
model,input, andvoicefields - Supports
tts-1andtts-1-hdmodels with identical engine performance - Uses standard
Bearertoken authentication for easy header management
Multi-Format Output (WAV, FLAC, OPUS)
Choose wav, flac, opus, mp3, or pcm to match your dataset requirements. Lossless formats like WAV and FLAC preserve audio fidelity for high-quality ML training, while OPUS offers efficient compression for large-scale data collection.
- WAV and FLAC for lossless, pristine audio data
- OPUS for balanced quality and storage efficiency
- PCM for raw, uncompressed byte streams
ISO Language Support
Specify the language parameter using ISO codes to ensure accurate synthesis, especially for Latin-script texts that might otherwise default to English. We support 23 languages, ensuring your multilingual datasets maintain consistent voice quality across diverse linguistic inputs.
- Explicit language codes for precise control (e.g.,
es,fr,ja) - Automatic script detection if
languageis omitted - Supports all 23 listed languages with any voice preset
23 Languages, Any Voice
Every voice supports all 23 languages, letting you build truly multilingual datasets without managing separate voice models. This eliminates the need to train language-specific acoustic models, streamlining your data collection workflow for global ML applications.
If the language isn't detected, set the language parameter explicitly. Text in Latin script defaults to English, so non-English Latin text requires a manual override to ensure accurate pronunciation and phoneme generation.
Arabic, Chinese, French, German
These languages cover major global markets. Use any voice preset to generate consistent synthetic speech for training multilingual ASR or TTS models without worrying about speaker variability.
Standard ISO codes are supported. Ensure your text matches the language script for optimal detection, or pass the language field to force the correct phonetic engine.
- Arabic, Chinese, French, German
- Any voice preset works
- ISO language code support
Japanese, Korean, Spanish, Russian
High-demand languages for international datasets. Generate thousands of hours of training data with consistent quality. Each request returns a clean audio file ready for immediate use in model pipelines.
Speed control allows you to adjust generation rates, ensuring the synthetic data matches the pacing of real-world speech for better model generalization.
- Japanese, Korean, Spanish, Russian
- Consistent voice quality
- Speed adjustment (0.25–4.0)
Scandinavian & Southeast Asian Variants
Support for Danish, Norwegian, Swedish, Finnish, Malay, and Turkish allows for niche dataset creation. These languages are often underrepresented in public datasets, giving your models a competitive edge.
Generate diverse acoustic samples quickly. The API handles complex phonemes and tones, delivering structured audio files that are ready for labeling and training.
- Danish, Norwegian, Swedish, Finnish
- Malay, Turkish
- Underrepresented language support
Pricing for High Volume
For ML dataset production, character count is the only metric that matters. Our API charges exactly $0.01 per 1,000 characters, including spaces and punctuation. You receive a precise, transparent invoice for every audio file generated, with no hidden fees for format conversions or speed adjustments.
$0.01 per 1k Characters
Each request processes up to 4,096 characters. For longer texts, split the input and join the resulting audio files. Failed requests are free, so you only pay for successful synthesis. This linear pricing makes budgeting for large-scale dataset creation predictable and simple.
Pay-Per-Use Model
Top up via USDT (TRC-20) on the TRON network. Credits are available immediately after the transfer confirms, typically within minutes. There are no monthly subscriptions or auto-renewing plans. Your balance does not expire, allowing you to pause and resume dataset generation at your own pace.
- Minimum top-up is $5.
- No VAT or invoices required.
No Minimum Spend
You can start with the free trial credit of $0.50 (about 50,000 characters). This is ideal for testing voice presets or validating your pipeline before committing to a larger top-up. There is no minimum spend to keep your account active.
Dataset Quality & Watermarks
For ML datasets, consistency is critical. Every generated audio file carries an inaudible watermark, clearly marking it as synthetic speech without affecting acoustic properties. This ensures your training data remains distinct from human recordings, preventing label noise in ASR or voice cloning models.
Inaudible Watermarking
The watermark is embedded directly into the audio waveform. It is transparent to standard speech models but provides a reliable metadata signal for dataset curation. You get clean, labeled synthetic data without manual tagging overhead.
Public Domain Sources
Our voices are derived from LibriTTS-R (CC BY 4.0) and LibriVox public-domain audiobook readings. This means you inherit clear copyright statuses for your commercial or research datasets. No hidden licensing fees or complex rights management for the underlying voice performances.
Clean Audio Output
Output formats like WAV, FLAC, and OPUS provide high-fidelity results. The API returns raw audio files ready for immediate ingestion into your data pipelines. No background noise or artifacts from third-party processing ensure consistent quality across thousands of samples.
API Integration
Send text to our POST endpoint and receive audio instantly. The request shape mirrors the OpenAI standard, allowing you to swap the base_url and use existing SDKs without rewriting your integration logic.
Simple JSON Endpoint
Submit a JSON body with model, input, voice, and response_format. You can specify formats like mp3, wav, opus, flac, or pcm directly in the payload.
- Endpoint: POST https://api.play09bet.com/v1/audio/speech
- Auth: Bearer token in header
- Max input: 4096 characters
Speed Control
Adjust the speed parameter between 0.25 and 4.0 to fit your dataset’s rhythm. This lets you generate varied speech rates for robust ASR training without manual post-processing.
Error Handling
Failed requests return zero charges, ensuring your billing stays accurate even if the queue is busy. Use standard HTTP status codes to detect synthesis failures and retry or log as needed.
Use Cases in AI
Synthetic speech powers modern AI pipelines by providing labeled, noise-free audio at scale. Unlike manual recording, our API generates consistent, watermark-marked files instantly, eliminating label noise and ensuring uniform voice quality across massive datasets.
ASR Model Training
Train automatic speech recognition models with diverse, high-quality synthetic audio. Generate thousands of variations across 23 languages to improve transcription accuracy without the cost of human annotators.
- Eliminate label noise for cleaner training data
- Scale data collection instantly via API
Voice Cloning Datasets
Build robust voice cloning models using our 8 licensed presets. Since we don't offer custom voice cloning, these diverse, consistent voices serve as ideal reference data for training third-party cloning algorithms.
- Consistent voice quality across all samples
- Watermarked synthetic speech for dataset integrity
Emotion Detection Data
While we don't offer explicit emotion tags, you can simulate emotional variance by adjusting speed parameters. Combine different voices (e.g., energetic Felix vs. calm Sage) with varying speeds to create nuanced datasets for emotion classification models.

Get Started Today
Start building your synthetic speech dataset immediately with $0.50 in free trial credit, valid for 7 days. This allows you to generate approximately 50,000 characters of audio to validate voice quality and format compatibility before committing to volume.
Create your account directly on our platform to receive your unique API key instantly. Use this key to authenticate requests to the OpenAI-compatible endpoint, ensuring seamless integration with your existing ML pipelines without waiting for manual approvals.
Access comprehensive documentation detailing the JSON request structure, available voices, and rate limits. Review the curl examples and SDK configurations to start generating structured audio files for ASR training or voice cloning projects within minutes.
Free Trial Credit
Start building your synthetic speech dataset immediately with $0.50 in free trial credit, valid for 7 days. This allows you to generate approximately 50,000 characters of audio to validate voice quality and format compatibility before committing to volume.
- $0.50 credit included automatically
- Valid for 7 days
- Approx. 50,000 characters
Instant API Key
Create your account directly on our platform to receive your unique API key instantly. Use this key to authenticate requests to the OpenAI-compatible endpoint, ensuring seamless integration with your existing ML pipelines without waiting for manual approvals.
Documentation
Access comprehensive documentation detailing the JSON request structure, available voices, and rate limits. Review the curl examples and SDK configurations to start generating structured audio files for ASR training or voice cloning projects within minutes.
- JSON body structure
- OpenAI SDK compatible
- cURL examples provided
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. These diverse synthetic datasets will help you train your custom voice models effectively.
23 languages
Any voice speaks any of these languages.
