Frequently asked questions
What is the most important factor?
Reliability and clear audio quality are key. VoiceGrid provides a stable API endpoint that returns consistent results. With 23 languages and 8 distinct voice presets, you can match the tone of your chatbot precisely. The engine handles character limits up to 4096, ensuring your bot delivers clear responses without complex infrastructure management.
Which voice is best for chatbots?
For chatbots, milo offers a friendly, conversational tone, while sage provides a calm, instructional style. Nova is bright and upbeat for engaging interactions. Atlas is neutral for news-style updates. You can test all eight presets via the /v1/voices endpoint to find the voice that best fits your application's personality.
Is OpenAI SDK better than REST?
It depends on your existing stack. VoiceGrid’s API matches the OpenAI speech endpoint shape exactly. You can use the official OpenAI SDK by simply changing the base_url to https://api.playh95.com/v1. This allows you to leverage familiar code structures without learning a new library, while maintaining full control over request parameters.
How do I test before buying?
Sign up with Google to receive a $0.50 free trial credit, valid for 7 days. This covers approximately 50,000 characters. You can generate audio files using the curl example on the /get-api-key/ page to verify voice quality, speed, and format support before committing to a USDT top-up for larger usage.
What is the best audio format?
Choose based on your client’s capabilities. MP3 is widely supported and compact. WAV offers uncompressed quality for high-fidelity needs. OPUS and FLAC provide efficient compression. PCM is raw audio. All formats are available via the response_format parameter. Select the format that balances quality with bandwidth efficiency for your specific chatbot deployment.
How do I handle rate limits?
The limit is 60 requests per minute per key. If you exceed this, requests may queue. For longer texts, split input into 4096-character chunks and join the resulting audio files. Monitor your usage dashboard. If you need higher throughput, you can issue a new API key to reset limits, as each key has independent quotas.
Is synthetic speech acceptable?
Yes, but users should know it is AI-generated. Every audio file includes an inaudible watermark marking it as synthetic speech. This is standard for TTS APIs. Ensure your chatbot’s personality aligns with the voice choice. Transparency builds trust, especially if your users are sensitive to artificial tones in customer service interactions.
How do I ensure data privacy?
Input text is used only to produce audio and is not used for training models. Generated audio is returned in the response and not stored on our servers. Request metadata like time and character count is kept solely for billing purposes. This ensures your conversation data remains private and transient during processing.
What if I need more languages?
VoiceGrid supports 23 languages including English, Spanish, French, German, Japanese, and Hindi. Any voice can speak any supported language. If the language is not set, it guesses from the script, defaulting to English for Latin text. Explicitly set the language parameter for non-English Latin scripts to ensure accurate pronunciation.
How do I scale my usage?
Scale by splitting long texts into 4096-character segments and concatenating the audio outputs. Use multiple API keys if you need to bypass the 60 requests per minute limit. Top up your balance with USDT (TRC-20) as needed. There are no monthly plans or seats, so you only pay for the characters you generate.
What are common pitfalls?
A common pitfall is exceeding the 4096 character limit per request, which causes errors. Another is ignoring rate limits, leading to queued requests. Ensure you set the language parameter for non-English text to avoid incorrect voice selection. Also, remember that failed requests are free, but successful ones incur charges, so handle errors gracefully.
How do I compare APIs?
Compare based on voice variety, language support, and pricing. VoiceGrid offers 8 voices and 23 languages at $10 per million characters. Check if the API shape matches your SDK. Verify if streaming or SSML is needed, as VoiceGrid does not offer these. Compare the ease of integration and the transparency of billing.
Is a free trial enough?
The free trial provides $0.50 credit, enough for about 50,000 characters. This is sufficient for testing voice quality and integration but not for production usage. For ongoing needs, top up with USDT. The trial is valid for 7 days, so plan your testing phase accordingly to ensure you can evaluate the service effectively.
How do I calculate costs?
Cost is $10 per 1 million characters, or $0.01 per 1,000 characters. Spaces and punctuation count as characters. Failed requests are free. Check your balance and usage on the /get-api-key/ page. Multiply your expected monthly character volume by $0.00001 to estimate costs. There are no hidden fees or subscription charges.
Voice quality and variety
VoiceGrid offers eight distinct voice presets engineered for chatbot interactions, ranging from the warm clarity of Aria to the energetic promo style of Felix. You can evaluate each voice using the dedicated endpoints to ensure the tone matches your application’s personality before committing to an integration.
Listen to voice presets
Browse the available voices via the /v1/voices endpoint. Each preset, such as Luna’s soft storytelling or Atlas’s neutral news style, is designed for specific conversational contexts. You can test them directly with a quick POST request to generate sample audio.
- Aria: warm, clear narrator
- Nova: bright, upbeat
- Luna: soft, storytelling
- Sage: calm, instructional
- Felix: energetic, promo
- Milo: friendly, conversational
- Atlas: neutral, news
- Orion: deep, documentary
Check for natural tone
The voices are derived from licensed recordings, ensuring a natural, human-like quality without the uncanny valley effect often found in robotic TTS. The engine processes text in real-time, maintaining consistent pacing and intonation suitable for voice assistants.
Assess emotional range
While the system does not support explicit emotion tags, the inherent character of each voice provides distinct tonal qualities. For instance, Milo offers a friendly conversational tone, while Orion delivers a deep, documentary-style narration, allowing you to select the best fit for your brand's voice.
Language support
Choosing the right voice engine depends heavily on linguistic coverage. VoiceGrid supports 23 languages, allowing you to reach a broad audience without juggling multiple providers. This range covers major markets like English, Spanish, German, and Japanese, ensuring your chatbot feels native to diverse users.
Every voice can speak any of the 23 supported languages. This flexibility means you can pick the perfect tone—like the warm narrator 'Aria' or the deep documentary style 'Orion'—regardless of the target language, maintaining a unified brand voice across borders.
Number of languages
VoiceGrid supports 23 languages, including Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, and Turkish. Notable exclusions are Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, and Bengali.
This specific list ensures focused quality rather than broad, shallow coverage. You get reliable, licensed voices for each supported language without the complexity of managing dozens of marginal dialects.
- 23 supported languages
- Excludes Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, Bengali
Quality of non-English voices
Every voice preset can speak any of the 23 languages. This means you aren’t forced to use a specific voice for a specific language. You can choose the tone that fits your chatbot’s personality—whether it’s the bright 'Nova' or the calm 'Sage'—and apply it to English, Spanish, or Japanese text.
The voices are based on licensed recordings from LibriTTS-R (CC BY 4.0) and LibriVox public-domain audiobook readings. This consistency helps maintain user trust across different linguistic contexts.
Auto-detection accuracy
If you do not set the language field, the API guesses the language from the script. Latin script defaults to English, which works for English text but may misidentify other Latin-script languages like Spanish or French.
For non-English Latin-script text, always set the language parameter explicitly. This ensures accurate pronunciation and prosody, avoiding awkward pauses or incorrect accents that can break the immersion in your chatbot experience.

API ease of use
VoiceGrid simplifies integration by mirroring the familiar OpenAI speech endpoint structure. You send a POST request to /v1/audio/speech with a JSON body containing your model, input text, and voice preset. This design means you can reuse existing client code or swap the base URL in the official OpenAI SDK without rewriting your logic.
The API supports standard formats like MP3, WAV, Opus, FLAC, and PCM. You control the output via simple parameters, making it easy to adjust file types for your specific chatbot constraints without needing complex conversion pipelines.
Simple endpoints
The core endpoint is straightforward: POST to /v1/audio/speech. Pass your text in the input field, select a voice from the 8 presets, and specify the model (tts-1 or tts-1-hd). The API returns the audio file directly in the response, ready for immediate playback in your app.
You can also retrieve available voices and models via GET /v1/voices and GET /v1/models. This transparency helps you verify supported languages and formats before building your integration.
- Single endpoint for generation
- Standard JSON request body
- Direct audio response
SDK availability
We do not provide a proprietary SDK. Instead, we rely on the widely used OpenAI SDKs. By simply changing the base_url to https://api.playh95.com/v1, you can use the official OpenAI libraries with your VoiceGrid key. This ensures compatibility with existing developer workflows and tools.
This approach avoids locking you into a specific vendor’s library while leveraging robust, well-documented client libraries that many developers already know.
Documentation clarity
Our docs clearly list all 23 supported languages and 8 voice presets. You’ll find exact field definitions for model, input, voice, response_format, and speed. The documentation also explains authentication via Bearer tokens and rate limits (60 requests/minute).
A ready-to-use cURL example is provided on the /get-api-key/ page, allowing you to test the API immediately after signing up. The docs specify that failed requests are free, reducing risk during development.
Pricing structure
VoiceGrid charges $10 per million characters, making it easy to predict costs for your chatbot’s voice output. Spaces and punctuation count as characters, so a 1,000-character request costs just $0.01. Failed requests are free, with the charge returned in full if synthesis fails, ensuring you only pay for successful audio generation.
Pay per character
The pricing is straightforward: $10 per million characters. This linear model scales naturally with your chatbot’s usage, avoiding complex tiered structures. You can estimate costs precisely based on your expected character volume, ensuring transparency in your development budget.
- $10 per 1,000,000 characters
- $0.01 per 1,000 characters
- Spaces and punctuation count as characters
- Failed requests are free
Free trial availability
New accounts can start with a $0.50 trial credit, valid for 7 days, allowing you to generate approximately 50,000 characters. This trial is automatically granted when you sign in with Google. Accounts created with an email and password begin without this initial credit, so choose your signup method wisely.
- $0.50 trial credit
- Valid for 7 days
- Automatic with Google sign-in
- Not included with email/password accounts
Top-up convenience
Add funds via USDT on the TRON network (TRC-20) with a minimum top-up of $5. Credits are added automatically within minutes of transfer, keeping your bot running without delay. There are no monthly plans, subscriptions, or VAT invoices, just a simple balance that never expires.
- USDT on TRON (TRC-20)
- Minimum $5 top-up
- Automatic credit within minutes
- No monthly plans or subscriptions
Performance and speed
Understanding generation speed is critical for chatbot responsiveness. Short phrases typically require a few seconds to synthesize, while longer texts generate at roughly real-time speed, averaging about 25 characters per second. This ensures your voice assistant feels natural without excessive waiting.
Generation time
Expect a few seconds of latency for short phrases. For longer inputs, the processing rate stabilizes at approximately 25 characters of text per second of audio generated, effectively matching real-time playback speed for most conversational use cases.
Rate limits
Your API key supports 60 requests per minute under normal conditions. New accounts start with a lower limit of 10 requests per minute until the first top-up is processed. This structure prevents sudden traffic spikes from overwhelming the service.
Queue handling
Requests are processed sequentially. Under high load, your synthesis job may wait in a queue before processing begins. This ensures fair resource allocation but means concurrent requests might experience slight delays during peak usage periods.
Integration requirements
Choosing the right integration path depends on your chatbot's architecture. VoiceGrid provides a straightforward HTTP API that fits easily into existing stacks, avoiding the complexity of managing WebSocket connections if you only need static audio responses.
This simplicity allows indie developers to focus on conversation logic rather than infrastructure. The RESTful design ensures reliable request handling without requiring persistent connections.
HTTP vs WebSocket ['VoiceGrid uses a simple POST endpoint for speech synthesis. You send text and receive an audio file in the response. This synchronous model is ideal for chatbots that generate a complete response before replying, avoiding the overhead of maintaining open WebSocket connections for every turn.', 'If your bot requires real-time streaming, you will need to implement chunking or polling, as the API does not support native WebSocket streaming. For most conversational flows, the standard HTTP request/response cycle is sufficient and easier to debug.'] []
SDK compatibility ['The API mirrors the OpenAI speech endpoint structure. This means you can use the official OpenAI SDKs by simply changing the base_url to https://api.playh95.com/v1. This compatibility allows you to leverage familiar libraries in Python, JavaScript, and Go without learning a new interface.', 'For simpler needs, standard HTTP libraries work perfectly. You can send JSON payloads with your text, voice selection, and format preference using any common HTTP client.'] []
Authentication method ['Authentication is handled via a Bearer token. You pass your API key in the Authorization header as Bearer tts-.... Each account holds one active key, which you can regenerate at any time if needed.', 'This simple key-based system integrates easily into environment variables or secret managers. It works seamlessly with both server-side backends and client-side applications that can safely expose the key.'] []
Data privacy policies
When building a chatbot, understanding where your user data goes is crucial for trust and compliance. Our API treats input and output data with strict isolation, ensuring your conversations remain yours. We do not repurpose your content for model improvement or share it with third parties without your explicit consent.
This approach allows indie developers to integrate TTS into sensitive applications, from customer support bots to health assistants, without worrying about accidental data leakage or unexpected usage of their unique voice scripts for training other models.
Input text usage
Your input text is used solely to produce the audio file requested. We do not store the original text after synthesis is complete, ensuring that your chatbot scripts remain private and are not accessible for other uses.
The generated audio is returned directly in the API response and is not stored on our servers. This means your user’s voice output exists only in the moment of playback, reducing your data retention burden.
- Input text is ephemeral
- Output audio is not stored
- No third-party sharing
Audio storage
Since we do not store generated audio, you are responsible for caching or delivering the audio files to your users. This gives you full control over your own storage infrastructure and ensures that your audio assets are not tied to our uptime or storage policies.
Request metadata, such as the timestamp, character count, and status, is kept briefly to facilitate accurate billing and troubleshooting. This minimal data is sufficient for invoice generation but does not include the audio content itself.
- Audio is not stored by us
- You control caching
- Metadata kept for billing only
Training data usage
Your input text is not used for training our TTS models. The voices are based on licensed recordings from LibriTTS-R and LibriVox, meaning your unique chatbot scripts do not contribute to the voice engine’s future iterations.
This separation ensures that if you have proprietary phrases or brand-specific scripts, they remain distinct from the general training data. Your data privacy is respected, and your content does not inadvertently improve competitors' models or alter the base voice quality.
- Text not used for training
- Proprietary scripts stay private
- No model contamination

Audio format options
VoiceGrid offers MP3, WAV, OPUS, FLAC, and PCM formats, letting you balance file size and fidelity for your chatbot’s audio assets.
Choose MP3 for efficient storage and streaming, or WAV/FLAC for lossless quality when audio fidelity is critical for your user experience.
MP3 vs WAV
MP3 files are smaller and ideal for bandwidth-constrained environments, while WAV provides uncompressed audio for maximum fidelity without data loss.
Select the format that best matches your chatbot’s delivery method, whether prioritizing quick loading or pristine sound reproduction.
- MP3: Compressed, smaller size, widely supported.
- WAV: Uncompressed, larger size, high quality.
- OPUS/FLAC/PCM: Additional options for specific needs.
File size vs quality
Larger files like WAV retain more audio detail but consume more storage and bandwidth compared to compressed formats like MP3.
Use FLAC for lossless compression if you need high quality without the full size of WAV, or PCM for raw data.
Balancing these factors ensures your chatbot remains responsive while delivering clear, understandable speech to users.
Streaming support
VoiceGrid does not offer real-time streaming or WebSocket endpoints; audio is generated and returned as a complete file.
You must handle the full file download or storage before playback, which simplifies integration but lacks live audio delivery.
Customer support
VoiceGrid relies on email for all support, keeping things simple and traceable without the overhead of phone lines or live chat agents. You can reach the team directly for any account or API questions.
Email support quality
Our support is handled entirely via email, ensuring every query is logged and answered by knowledgeable engineers rather than generic agents. There is no phone number or live chat widget to distract from the core API experience.
Response time
While we do not publish a strict SLA, our team prioritizes developer inquiries efficiently. You can expect prompt replies to your emails without waiting on hold or navigating complex ticket queues.
Documentation updates
The documentation is kept current with every API change, ensuring you always have accurate endpoint details, code examples, and pricing tables. We update the guides directly to reflect the latest capabilities of the VoiceGrid engine.
Scalability
Scaling your chatbot’s voice layer is straightforward because VoiceGrid operates on a simple, predictable model. There are no monthly plans or reserved seats to manage, allowing you to handle increased traffic without complex infrastructure adjustments. You simply consume characters as needed, ensuring your voice assistant grows alongside your user base without administrative overhead.
Handle increased traffic
Requests are processed sequentially, meaning under heavy load a request may wait in a queue. However, the engine handles long texts by splitting them into 4096-character parts and joining the audio files seamlessly. This ensures consistent delivery even when your chatbot processes multiple user interactions simultaneously.
Rate limit flexibility
Each API key allows 60 requests per minute once funded, with a lower 10 requests per minute limit during the free trial. Since there are no minimum spends or tiered plans, you can instantly increase capacity by topping up your USDT balance, which is credited automatically within minutes.
- 60 requests/minute for standard accounts
- 10 requests/minute for new trial accounts
- No hard caps on total volume
Cost scaling
Costs scale linearly with usage at $0.01 per 1,000 characters. Failed requests are free, so you only pay for successful synthesis. Your balance does not expire, allowing you to accumulate credits for peak usage periods without worrying about monthly plan renewals or unused seat costs.
Technical limitations
Understanding the technical boundaries ensures your chatbot handles edge cases gracefully without unexpected failures or delays.
Knowing these limits helps you design robust fallbacks and optimize text length before sending requests.
Character limits
Each API request accepts a maximum of 4096 characters. This includes spaces and punctuation. For longer scripts, split the text into chunks and join the resulting audio files.
- Max 4096 characters per request
- Spaces and punctuation count toward the limit
Speed range
You can adjust the playback speed between 0.25 and 4.0. This allows fine-tuning the voice to match your chatbot’s personality, from slow and clear to fast-paced.
- Speed range: 0.25x to 4.0x
- Adjustable per request
Format restrictions
Supported output formats are mp3, wav, opus, flac, and pcm. Choose based on your bandwidth needs and player compatibility. We do not support streaming or WebSocket audio delivery.
- Formats: mp3, wav, opus, flac, pcm
- No real-time streaming support

Decision checklist
Before committing to an API, ensure it aligns with your chatbot's specific conversational needs. Evaluate whether the voice presets match your brand's tone and if the language support covers your target audience.
Define use case
Identify if your bot requires fast, concise responses or longer, narrative-style interactions. Verify that the 4096-character limit per request fits your dialogue flow and that the speed range (0.25–4.0) allows for natural pacing without sounding robotic.
Test with demo
Use the $0.50 free trial to generate actual dialogues. Check if the auto-detection for Latin-script text works correctly and if the audio quality remains clear under typical load conditions.
- Test multiple voices like 'nova' for upbeat interactions or 'atlas' for news-style updates.
Check pricing fit
Calculate costs based on $0.01 per 1,000 characters. Ensure your expected monthly volume fits the $10 per million characters rate and that the TRC-20 top-up method suits your payment preferences.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. Design conversational experiences that feel natural for users interacting with voice assistants and chatbots daily.
23 languages
Any voice speaks any of these languages.
Sources and further reading
- Wikipedia
- Node.js documentation
- IETF RFC 7845: Ogg Opus
- LibriVox public-domain audiobooks
- TRON developer hub: TRC-20
- FFmpeg documentation
- Python documentation
- curl documentation
- W3C WAI: making audio and video accessible
- ISO 639 language codes
- OpenAI API reference: create speech
- Creative Commons Attribution 4.0
