Comprehensive Guide to API Voices, Formats, and Languages

By PlayVeBet editorial team · Updated 2026-11-18

This guide details the voice presets, audio formats, and language support for the PlayVeBet TTS API. It is designed for engineering teams at language learning startups who need to integrate synthetic speech into educational workflows. You will find specific technical specifications for all eight voice models, including tone distinctions between female and male presets.

We explain how to manage audio generation constraints, such as the 4096-character limit, speed controls, and rate limits. The document also covers licensing details for LibriTTS-R sources and clarifies unsupported features like SSML. Use this reference to configure your API requests accurately and ensure character-accurate speech output for your application.

Comprehensive Guide to API Voices, Formats, and Languages

Frequently asked questions

What is the difference between tts-1 and tts-1-hd?

Both models use the same engine and have identical pricing. The distinction lies only in the model identifier sent in the request body. Select "tts-1" or "tts-1-hd" based on your preference, but expect the same audio quality and cost per character for either selection.

Can I use Aria for storytelling content?

Yes. Aria is described as a warm, clear narrator voice. This preset is suitable for storytelling content where a clear, engaging delivery is required. You can select Aria via the voice parameter in your POST request to generate audio files for your educational applications.

Is FLAC better than MP3 for learning apps?

FLAC and MP3 are both supported response formats. FLAC offers lossless compression, while MP3 is lossy. The API does not specify technical superiority for learning contexts. Choose the format that best fits your application's storage and bandwidth constraints, as both are generated from the same engine.

Which voice is best for instructional videos?

Sage is described as a calm, instructional voice. You can select Sage via the voice parameter in your POST request to generate audio files for your educational applications.

How do I specify Japanese in the request?

Include the "language" field in your JSON request body with the ISO code "ja". Although the API can guess language from the script, explicitly setting the language parameter ensures accurate synthesis for Japanese text, preventing potential misinterpretation by the engine during the TTS process.

Does the API support Vietnamese yet?

No. Vietnamese is explicitly listed as not supported. The API supports 23 languages, including English, French, and German, but Vietnamese is excluded. Ensure your application filters or handles Vietnamese text separately before sending requests to the API to avoid errors or unexpected results.

What is the maximum speed setting?

The maximum speed setting is 4.0. You can adjust the speed parameter in your request between 0.25 and 4.0. This allows you to control the playback rate of the generated audio file to suit different educational pacing requirements without altering the voice characteristics.

Are the voices based on real people?

Yes. The voices are based on licensed recordings from LibriTTS-R (CC BY 4.0) and LibriVox public-domain audiobook readings. Every generated file carries an inaudible watermark that marks it as synthetic speech.

Can I detect the language automatically?

Yes. If the "language" field is omitted, the API guesses the language from the script. Latin script text defaults to English. For non-English Latin-script text, you should explicitly set the language parameter to ensure accurate synthesis, as automatic detection may not always be precise.

How long is my text stored?

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. Request metadata, including time, characters, and status, is kept for billing purposes. Your text is not retained for other uses.

Is there a limit to how many voices I can use?

There is no specified limit to the number of voices you can use. You can select from eight presets (Aria, Nova, Luna, Sage, Felix, Milo, Atlas, Orion) in your requests. Each voice can be used repeatedly across different requests without restriction based on the account or key.

Do I need to pay for SSML tags?

The API does not support SSML. There are no SSML tags to pay for or use. The API accepts plain text input. You cannot include SSML markup in the input field. If you require SSML, you must handle that functionality outside of this API service.

What happens if I send 5000 characters?

The API accepts up to 4096 characters per request. If you send 5000 characters, you must split the text into parts of up to 4096 characters. Generate audio for each part and join the resulting audio files. The API does not automatically split long texts for you.

Is the API suitable for commercial use?

Yes. The API provides a standard HTTP endpoint for generating audio files. You can integrate it into your applications for commercial purposes. The voices are based on licensed recordings, and generated audio is returned for your use.

Voice Preset Library Overview

Sound wave illustration for: Voice Preset Library Overview

PlayVeBet offers eight distinct voice presets engineered for clarity and consistency in language learning applications. The library is divided into female and male categories, ensuring that educational content can be delivered with appropriate tonal variations. Each voice is designed to maintain character accuracy across diverse linguistic contexts.

The selection allows developers to assign specific vocal identities to different learning modules or characters. By leveraging licensed recordings from LibriVox, the voices provide a natural, human-like cadence that supports effective language acquisition without the need for custom voice design.

Female Voice Characteristics

The female presets include Aria, Nova, Luna, and Sage. Aria provides a warm, clear narrator tone suitable for standard lessons. Nova offers a bright, upbeat style for engaging exercises. Luna delivers a soft, storytelling voice for narrative-based learning. Sage ensures a calm, instructional tone for structured content.

These voices cover a range of emotional tones from energetic to serene, allowing educators to match the voice to the lesson's intensity without requiring emotional tagging or custom cloning.

  • Aria: Warm, clear narrator
  • Nova: Bright, upbeat
  • Luna: Soft, storytelling
  • Sage: Calm, instructional

Male Voice Characteristics

The male presets consist of Felix, Milo, Atlas, and Orion. Felix is energetic and ideal for promotional or high-energy content. Milo provides a friendly, conversational tone for dialogue practice. Atlas offers a neutral, news-style delivery for factual content. Orion delivers a deep, documentary voice for immersive storytelling.

Each male voice is derived from licensed LibriTTS-R recordings, ensuring consistent audio quality and compliance with CC BY 4.0 standards while supporting the full range of 23 supported languages.

  • Felix: Energetic, promo
  • Milo: Friendly, conversational
  • Atlas: Neutral, news
  • Orion: Deep, documentary

Tone and Style Distinctions

The distinction between voices lies in their tonal range and pacing. Female voices range from the upbeat energy of Nova to the soft narrative flow of Luna. Male voices span from the conversational warmth of Milo to the authoritative depth of Orion.

These distinctions allow for precise control over the auditory experience in educational workflows. Each voice can speak any of the 23 supported languages, enabling consistent character voice across different linguistic modules without needing separate voice models for each language.

image":

Detailed Voice Profiles

Sound wave illustration for: Detailed Voice Profiles

The API provides eight distinct voice presets tailored for educational content. These voices are derived from licensed recordings, ensuring consistent quality across language learning applications. Each preset offers specific tonal characteristics suitable for different instructional contexts.

Startups can select voices based on the desired listener engagement level. The library includes both male and female options, allowing for diverse audio experiences in course materials.

Aria: Warm Narrator Tone

Aria delivers a warm and clear narration style. This preset is ideal for standard educational content where clarity is paramount. Its neutral yet engaging tone supports long-form listening sessions without causing listener fatigue.

Use Aria for general knowledge modules or vocabulary drills. The voice maintains a steady pace, ensuring that pronunciation remains accurate and easy to follow for language learners.

  • Warm, clear narrator
  • Suitable for general education
  • Steady pacing for clarity

Nova: Bright Upbeat Style

Nova offers a bright and upbeat delivery. This voice is designed to keep learners engaged during interactive exercises or quick quizzes. The energetic tone helps maintain attention during shorter audio segments.

Nova is effective for introductory segments or reinforcement activities. Its lively character can make repetitive practice feel less monotonous for students.

  • Bright, upbeat style
  • High engagement level
  • Ideal for quizzes

Luna: Soft Storytelling Voice

Luna provides a soft voice optimized for storytelling. This preset is best suited for narrative-based lessons or reading comprehension tasks. The gentle delivery enhances immersion in story-driven content.

Use Luna for parables, fables, or conversational examples. The tone supports narrative flow in dialogue, helping learners understand context and flow in natural speech.

  • Soft, storytelling voice
  • Narrative focus
  • Emotional nuance

Audio Format Specifications

Sound wave illustration for: Audio Format Specifications

PlayVeBet’s TTS API supports four standard audio formats: MP3, WAV, Opus, and FLAC. Each format serves distinct technical requirements for language learning applications, allowing developers to balance file size, compatibility, and audio fidelity based on their specific workflow needs.

MP3 Compression Standards

MP3 provides efficient lossy compression suitable for web delivery and mobile applications where bandwidth conservation is critical. This format ensures broad device compatibility while maintaining acceptable speech clarity for educational content playback.

  • Supports variable bitrate for optimized storage
  • Ideal for streaming and mobile apps

WAV Uncompressed Quality

WAV delivers uncompressed PCM audio, preserving the full fidelity of the synthesized voice without data loss. Note that the API also supports a separate 'pcm' response format. This format is optimal for local editing, archival storage, or integration with audio processing pipelines that require raw audio data.

  • ['Lossless preservation of audio data', 'Best for local editing workflows']

FLAC Lossless Options

FLAC offers lossless compression, reducing file size significantly while retaining all original audio information. This format is ideal for high-quality archival needs or scenarios where future re-encoding might degrade quality, ensuring consistent educational standards over time.

  • ['Significant size reduction without quality loss', 'Suitable for high-fidelity archival']

Language Support Matrix

Sound wave illustration for: Language Support Matrix

PlayVeBet supports 23 languages, enabling precise localization for language learning applications. The API handles 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 preset can render any of these languages with consistent quality.

Accurate synthesis relies on correct language identification. The API uses script-based detection for Latin scripts, defaulting to English if unspecified. For non-English Latin-script text, explicitly setting the language field is required. For non-English Latin-script text, explicitly setting the language field is required. Languages such as Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, and Bengali are not supported.

23 Supported Languages List

The API covers 23 languages, ensuring broad coverage for educational content. Supported languages include Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, and Turkish.

None

  • Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish

Script-Based Detection Logic

The system detects language from the input script. Latin script text defaults to English if no language is specified. This automatic detection simplifies integration for English-centric workflows.

None

Explicit Language Code Usage

For non-English Latin-script text, you must set the language field to the ISO code. This ensures accurate pronunciation and avoids misidentification. Unsupported languages include Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, and Bengali.

None

Speed and Timing Controls

Sound wave illustration for: Speed and Timing Controls

The API supports a speed multiplier ranging from 0.25x to 4.0x, allowing precise control over speech pacing for educational contexts. This flexibility ensures that synthetic voices can be adjusted to match the processing speed of language learners without requiring post-processing.

0.25x to 4.0x Range

Users can specify any decimal value within the 0.25 to 4.0 range. Lower speeds aid comprehension for beginners, while higher speeds accelerate content delivery for advanced users. The API handles all temporal adjustments during synthesis, ensuring consistent audio quality across different paces.

  • Minimum speed: 0.25x (quarter speed)
  • Maximum speed: 4.0x (quadruple speed)
  • Default speed: 1.0x

Real-Time Generation Rate

Synthesis typically processes text at roughly real-time speed, generating approximately 25 characters of text per second of audio. For short phrases, generation may take a few seconds due to initialization overhead. This rate applies to standard processing conditions.

Queue Delay Impact

Requests are processed sequentially. Under high load, a request may wait in a queue before synthesis begins, increasing total latency beyond the base generation time. This sequential processing ensures consistent resource allocation but may cause variable response times during peak usage.

Character Limit Mechanics

Sound wave illustration for: Character Limit Mechanics

The API enforces a strict 4096-character limit per request, counting spaces and punctuation. This constraint applies to all model variants and response formats, requiring developers to manage input length programmatically for longer educational texts.

4096 Character Max

Each API call processes up to 4096 characters of input text. Requests exceeding this limit are rejected, ensuring consistent processing times and predictable billing calculations for language learning applications.

Splitting Strategies

Developers must implement client-side logic to segment longer passages into chunks of 4096 characters or fewer. This splitting should respect semantic boundaries, such as sentence endings, to maintain audio coherence when the parts are recombined.

Audio Joining Process

Once multiple audio files are generated from segmented text, they must be concatenated on the client side. The API returns each segment as a separate file; no server-side merging is performed, giving full control over the final audio structure.

Data Privacy and Ownership

Sound wave illustration for: Data Privacy and Ownership

Input text is processed solely to generate the requested audio file and is not used for model training. This ensures that your educational content remains proprietary to your application.

Generated audio files are returned directly in the API response and are not stored on our servers after delivery. Your data is transient, existing only for the duration of the request processing.

Input Text Non-Training Usage

The text you submit via the POST /v1/audio/speech endpoint is used exclusively for synthesis. We do not retain your input text for future training or improvement of our TTS engines, preserving the uniqueness of your language learning materials.

Audio File Non-Storage Policy

Once the audio file is delivered to your client, we do not store a copy on our infrastructure. This minimizes data residency risks and ensures that your generated speech assets are controlled entirely by your infrastructure.

  • Audio is returned in the response body.
  • No persistent storage of generated audio files.
  • You control the storage of downloaded audio.

Metadata Retention Period

We retain request metadata, including timestamp, character count, and status, strictly for billing and usage verification purposes. This data supports accurate charge calculation based on the $0.01 per 1,000 characters rate.

Data Privacy and Ownership

Model Variants Comparison

Sound wave illustration for: Model Variants Comparison

The API offers two model identifiers, "tts-1" and "tts-1-hd", which route requests to the identical synthesis engine. While the label suggests a quality difference, both variants generate audio using the same underlying processing logic and voice presets.

tts-1 Standard Quality

The "tts-1" model identifier serves as the standard entry point for speech generation. It processes input text through the core engine to produce audio files in the requested format, such as MP3 or WAV.

tts-1-hd Enhanced Quality

The "tts-1-hd" identifier utilizes the same engine architecture as the standard model. Despite the "HD" designation, it uses the same engine and price as the standard model.

Price and Engine Parity

Both models incur the same rate of $0.01 per 1,000 characters. There is no premium pricing tier for the "HD" variant, and switching between them yields no perceptible difference in audio fidelity or generation speed.

  • Identical engine architecture for both models
  • Same character-based pricing structure
  • No audible quality difference in output

Watermarking Technology

Sound wave illustration for: Watermarking Technology

Every audio file generated through the PlayVeBet API contains an inaudible digital watermark. This mark is embedded during synthesis to clearly indicate that the speech is synthetic, ensuring transparency in content origin without affecting the listening experience for end-users.

Inaudible Synthetic Mark

The watermark is imperceptible to human ears, preserving audio quality while maintaining a reliable signal. It is automatically applied to all outputs regardless of the selected voice or format, ensuring consistent detection across MP3, WAV, and other supported formats.

Detection Capability

Specialized algorithms can identify the presence of this synthetic mark in the audio file. This allows platforms to verify whether speech was generated by the API rather than recorded by a human voice actor, providing a layer of authenticity verification.

Content Identification

The mark serves as a clear identifier for synthetic speech, aiding in content management and rights verification. It helps language learning applications distinguish between user-recorded exercises and API-generated examples, ensuring accurate metadata tracking for educational workflows.

Rate Limit Implications

Sound wave illustration for: Rate Limit Implications

Rate limits define the throughput capacity for your language learning application. The API enforces a standard ceiling of 60 requests per minute per active key, ensuring stable service distribution across all network accounts. This limit applies to the POST /v1/audio/speech endpoint and is strictly monitored to prevent resource exhaustion during peak educational hours.

60 Requests Per Minute

Once your account is established and funded, you may submit up to 60 synthesis requests within a single minute. This throughput supports moderate-volume educational workflows, such as generating lesson audio for multiple student groups simultaneously. Exceeding this threshold results in immediate HTTP 429 errors until the window resets.

Initial 10 Request Limit

New accounts begin with a reduced capacity of 10 requests per minute. This constraint remains in effect until the first top-up is processed via USDT (TRC-20). The lower initial limit protects infrastructure during the onboarding phase, requiring new integrations to implement basic retry logic with exponential backoff.

Load Balancing Effects

Requests are processed sequentially within the available quota. Under heavy load, a request may wait in a queue before synthesis begins. This serialization ensures consistent audio quality but can increase total generation time for large batches of text. Applications should account for this potential latency when scheduling audio delivery to students.

Voice Source Licensing

Sound wave illustration for: Voice Source Licensing

PlayVeBet’s synthetic voice engine is constructed from licensed recordings sourced directly from the LibriTTS-R dataset. This foundation ensures that all generated audio stems from verified, high-quality human speech samples rather than generic statistical models. The voices are based on licensed recordings from the LibriTTS-R dataset, providing the acoustic characteristics that define each unique tonal profile.

LibriTTS-R Attribution

The underlying voice data originates from the LibriTTS-R dataset, which aggregates thousands of hours of read speech. By utilizing this specific dataset, we maintain a consistent baseline for voice quality across all our endpoints. The attribution ensures that the source material remains traceable and compliant with open licensing standards.

CC BY 4.0 Compliance

All source recordings are distributed under the Creative Commons Attribution 4.0 International license. This permits the commercial use of the audio data for generating synthetic speech without additional royalties. We handle the attribution requirements, allowing you to integrate the voices into language learning applications without managing individual licensing agreements for the source material.

LibriVox Source Material

The LibriTTS-R dataset is derived from public-domain audiobook readings recorded by volunteers on LibriVox. These recordings provide natural prosody and diverse speech patterns. Because the original content is public domain, the resulting synthetic voices can be deployed in educational contexts without fear of copyright infringement on the voice identity itself.

Unsupported Features List

Sound wave illustration for: Unsupported Features List

PlayVeBet provides a streamlined HTTP API focused on core text-to-speech functionality. We prioritize reliability and standard integration over complex feature sets. Understanding what is excluded helps you plan your language learning application architecture correctly.

No SSML Support

The API does not support Speech Synthesis Markup Language (SSML). You cannot use tags to control pauses, pitch, or volume within the input text. If you require precise prosodic control, you must manage timing externally or pre-process your text content before sending it to the endpoint.

No Custom Voice Cloning

Custom voice cloning is not offered. You cannot upload a sample recording to create a unique voice model. The eight available presets (Aria, Nova, Luna, Sage, Felix, Milo, Atlas, Orion) are fixed based on licensed recordings from LibriTTS-R and LibriVox. All generated audio carries an inaudible watermark indicating synthetic origin.

No Streaming Capability

There is no WebSocket or streaming support for partial audio chunks. The API returns the complete audio file only after synthesis finishes. For real-time playback in your language learning app, you must wait for the full response or implement client-side buffering strategies.

Listen to the 8 voices

Each sample below was generated by this API from the same sentence. Empower your global team with intuitive language learning tools that bridge cultural gaps and accelerate collaboration.

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. FFmpeg documentation
  3. IETF RFC 7845: Ogg Opus
  4. Python documentation
  5. LibriTTS-R corpus (OpenSLR 141)
  6. Creative Commons Attribution 4.0
  7. curl documentation
  8. ISO 639 language codes
  9. openai-python on GitHub
  10. Node.js documentation
  11. TRON developer hub: TRC-20
  12. OpenAI API reference: create speech

More on this site