Frequently asked questions
Are you affiliated with OpenAI?
No. We are an independent review site. Our API endpoint is https://api.marsgaming.info/v1, distinct from OpenAI’s infrastructure. We use the OpenAI SDK format for compatibility but are not affiliated with OpenAI, ElevenLabs, Google, Amazon, Microsoft, or any other brand mentioned in comparisons.
Who owns the voices?
The voices are based on licensed recordings from LibriTTS-R (CC BY 4.0) and public-domain audiobook readings from LibriVox. We do not own the original recordings but license them for our synthetic speech engine. Each generated file carries an inaudible watermark marking it as synthetic speech.
How do you verify the voices?
We analyze the eight presets (aria, nova, luna, sage, felix, milo, atlas, orion) for clarity, emotional range, and technical compatibility. We test them against gaming and voiceover workflows, verifying they support 23 languages and handle text up to 4096 characters per request.
Do you store user data?
Input text is used only to produce audio and is not used for training. Generated audio is returned in the response and not stored. We keep request metadata (time, characters, status) strictly for billing purposes. No personal data is retained beyond what is necessary for the API service.
Can I contact you by phone?
No. We do not have a phone number or live chat. Support is provided by email only. We maintain this lean support structure to keep our API pricing at $0.01 per 1,000 characters.
How often do you update reviews?
We update reviews when the underlying voice presets or API specifications change. Since we analyze fixed presets (aria, nova, etc.) and stable engine behavior, updates are infrequent. We verify current pricing ($10/1M characters) and rate limits (60 requests/minute) to ensure our data remains accurate for streamers.
Do you accept sponsorships?
We do not accept sponsorships that influence our technical reviews. Our analysis focuses on objective metrics like character limits, speed, and language support. We are not affiliated with the API provider, ensuring our recommendations for voices like felix or atlas remain unbiased and specific to creator needs.
What is your data retention policy?
Input text is processed and not stored for training. Generated audio is returned in the response and not stored on our servers. We retain request metadata (time, characters, status) only for billing and dispute resolution. This ensures minimal data footprint for our users and partners.
Can I suggest a voice for review?
We currently review the eight available presets: aria, nova, luna, sage, felix, milo, atlas, and orion. Since we do not offer custom voice cloning or design, we cannot add new voices. We focus on optimizing the usage of these eight existing synthetic voices for gaming content.
How do you define 'realistic'?
We define realism by natural prosody, clear articulation, and lack of robotic artifacts in the generated audio. We test this by generating long texts (up to 4096 characters) and checking for consistency. Each file includes an inaudible watermark, confirming synthetic origin without compromising the listening experience for viewers.
Do you test voices in real streams?
We test voices by generating audio files via the API and analyzing their playback quality in stream overlays and VODs. We verify that voices like nova or milo maintain clarity under typical stream conditions. We do not test live WebSocket streaming, as our API returns complete audio files.
Is the editorial team paid by the API provider?
No. We are an independent storefront. Our revenue comes from API usage, not editorial sponsorships. We maintain strict separation between our reviews of voices like aria or atlas and our commercial operations, ensuring unbiased advice for VTubers and content creators.
Can I use your reviews commercially?
Our reviews are factual analyses of the API’s capabilities. You may reference our data on voice characteristics and pricing. However, the generated audio files themselves carry a synthetic watermark. We do not claim ownership of the voice recordings, which are based on LibriVox public-domain readings.
How do you handle errors in reviews?
If we identify inaccuracies in our voice descriptions or pricing, we update the page immediately. We verify all claims against the live API endpoint. Failed synthesis requests are free, so we also test error handling to ensure our reviews reflect the actual reliability of the service for streamers.
What is your contact email?
Support is by email only. You can reach us via the contact page on marsgaming.info. We do not have a phone number or live chat. Email support is the primary channel for technical inquiries regarding the API, voice presets, or billing issues with your account.
Do you have a physical address?
We do not list a physical address. As a review site, we operate remotely. This allows us to maintain low overhead and offer competitive pricing ($0.01 per 1,000 characters) without the costs associated with physical office spaces or regional data centers.
Can I request a specific voice comparison?
Yes. You can request comparisons between any of the eight presets (e.g., felix vs. atlas). We generate audio samples using the API and analyze them for tone, speed, and clarity. We focus on how these voices perform in gaming contexts, such as commentary or character voices.
Do you test voice longevity?
We test longevity by generating long texts and splitting them into 4096-character chunks. We verify that the audio joins seamlessly and that the voice maintains consistent tone and quality over extended periods. This ensures reliable performance for long-form stream content or recorded videos.
How do you measure 'energy'?
We measure energy by analyzing the prosody, pitch variation, and pacing of voices like felix or nova. We generate promotional-style text and evaluate the perceived enthusiasm. This subjective metric is backed by objective tests of clarity and consistency across different languages and speeds.
Is the site mobile-friendly?
Yes, our site is optimized for mobile devices. We ensure that API documentation, pricing tables, and voice reviews are accessible on smartphones and tablets. This allows streamers to check voice specs and generate audio on the go, using the same endpoint: https://api.marsgaming.info/v1.
Who We Are
We are an editorial team dedicated to helping VTubers and streamers find a vocal identity for their content. Our reviews focus on clarity, emotional range, and technical compatibility for gaming and voiceover work, providing concrete examples and data-driven insights to guide your choice.
Each site in our network operates as an independent storefront for the same Text-to-Speech API. We maintain strict editorial independence, ensuring that our analysis of the eight available synthetic voices remains unbiased and specific to the needs of content creators.
Editorial Team Focus
Our team specializes in analyzing the eight presets—such as Aria, Nova, and Atlas—to determine their suitability for streaming workflows. We evaluate each voice based on production value, ensuring that creators can make informed decisions about their audio assets without guesswork.
Independent Review Process
We generate audio using the live API endpoint to verify response times, character limits, and voice characteristics firsthand. This hands-on approach allows us to report accurate data on speed, quality, and technical constraints directly relevant to your broadcast needs.
No Brand Affiliation
We are not affiliated with OpenAI, ElevenLabs, Google, Amazon, Microsoft, or any other brand mentioned in comparisons. Our reviews are based solely on the facts of the VoxHype API, ensuring that our recommendations are transparent and free from external brand influence.
Voice Licensing and Sources
The eight synthetic voices powering our API are derived from licensed recordings of the LibriTTS-R dataset, which carries a CC BY 4.0 license. This foundational dataset provides the vocal characteristics that define the distinct personalities of voices like Aria, Felix, and Orion.
Each voice model is trained on public domain audiobook readings sourced directly from LibriVox. By leveraging these freely available recordings, we maintain a transparent lineage for the audio synthesis engine, allowing content creators to trust the legal provenance of the synthetic speech they generate for their streams.
LibriTTS-R License (CC BY 4.0)
The underlying technology relies on the LibriTTS-R dataset, which is licensed under Creative Commons Attribution 4.0 International. This license permits commercial use and modification, provided appropriate credit is given to the original dataset creators. Our integration respects these terms while delivering a seamless API experience.
- Dataset: LibriTTS-R
- License: CC BY 4.0
- Allows commercial derivative works
LibriVox Public Domain Source
The raw audio inputs originate from LibriVox, a project providing public domain audiobooks. These recordings form the base for our voice presets, ensuring that the emotional range and clarity are grounded in established, high-quality human narration. This public domain source allows for flexible licensing of the resulting synthetic voices.
- Source: LibriVox audiobooks
- Status: Public Domain
- Base for all 8 voices
Commercial Use Rights
Users can generate audio for commercial purposes, including monetized streams and video content, without additional licensing fees beyond the standard per-character rate. The CC BY 4.0 license of the source data supports this commercial usage model, allowing streamers to integrate these voices directly into their content creation workflows.
- Commercial use allowed
- No extra licensing fees
- Suitable for monetized content
Data Privacy and Usage
Our API treats your input text as a transient payload. It is used exclusively to synthesize the requested audio file and is never ingested into our training datasets to improve the model. This ensures that your unique voice lines, stream scripts, and proprietary content remain private and do not influence future voice generations.
Input Text Not Used for Training
Unlike many consumer TTS services, we do not use your submitted text to train our models. Your input is processed solely to generate the immediate audio response and is discarded after that purpose is fulfilled, keeping your creative content distinct from our engine's development data.
Generated Audio Not Stored
The audio files returned in the API response are not stored on our servers. Once the synthesis is complete and the file is delivered to your client, the generated audio is removed from our infrastructure. This prevents any long-term accumulation of your voice assets in our database.
Request Metadata for Billing
We retain only the necessary request metadata to accurately calculate your usage and charges. This includes the timestamp, character count, and request status. This data is used strictly for billing purposes and to ensure you are charged correctly for the 4096-character limits and speed settings applied.
Technical Limitations
Our API delivers full audio files via standard HTTP responses, not partial streams. This means you receive the complete .mp3 or .wav file only after synthesis finishes. For real-time partial audio, you must implement your own buffering logic or wait for the full download.
We do not support SSML tags for precise pronunciation control or prosody adjustments. Speed and pitch are adjusted globally via the speed parameter (0.25–4.0), but you cannot mark pauses or emphasize specific words within a single request.
No Streaming or WebSocket
The endpoint returns the final audio file once processing completes. There is no WebSocket connection or chunked transfer encoding for live audio playback. This fits batch processing workflows but requires waiting for the full response.
Generation typically takes roughly real-time duration (about 25 characters per second of generation). Short phrases may take a few seconds, while longer texts queue behind other requests. You cannot interrupt or stream partial audio during synthesis.
- HTTP POST returns full file
- No WebSocket support
- No chunked streaming
No SSML Support
We use a simplified request shape compatible with the OpenAI SDK. You cannot inject SSML tags like <break> or <prosody> to control intonation or pauses. Complex prosody must be handled by splitting text into separate requests.
Speed adjustments apply to the entire input text uniformly. There is no phoneme dictionary or custom lexicon for specific pronunciation overrides.
- No <break> tags
- No prosody control
- Global speed only
No Voice Cloning Offered
We offer eight fixed presets based on licensed recordings. You cannot upload a sample voice to create a custom clone. There is no voice design tool or emotion tagging system available.
All generated files carry an inaudible synthetic watermark. This identifies the output as AI-generated but does not indicate which specific preset was used.
- 8 fixed voices
- No custom cloning
- No emotion tags
Contact and Support
We provide direct email support for all technical and billing inquiries related to the VoxHype TTS API. Our team handles account issues, API integration questions, and voice selection advice via a single dedicated channel.
We do not offer live chat, phone support, or on-site assistance. This keeps our overhead low and allows us to pass those savings directly to streamers and content creators using our API.
Email Support Only
All support requests are processed through our dedicated support email address. You can reach us at [email protected] for any issues regarding your API key, billing, or voice selection.
Our email system is optimized for clear, documented communication, ensuring that every query is tracked and resolved without the noise of live chat queues.
- Dedicated support email: [email protected]
- No third-party ticketing forms
No Live Chat or Phone
We intentionally do not maintain live chat widgets or a customer service phone line. This decision eliminates the need for large support teams, keeping our service affordable for independent creators.
You will not find a phone number on any of our pages, including /contact/ or /about/. All communication is asynchronous and documented.
Response Times
We aim to respond to all support emails within 24 hours on business days. During high-traffic periods, such as new voice releases or API updates, response times may extend slightly.
For urgent API key issues, we prioritize account access problems to ensure your streaming workflow remains uninterrupted.
Our Mission
We analyze the eight synthetic voices to help VTubers and streamers find the perfect vocal identity. Our goal is to eliminate guesswork by providing clear, factual comparisons of clarity, emotional range, and technical compatibility for gaming content.
Helping Streamers Choose Voices
We focus on how each voice performs in live streaming contexts. By testing aria, nova, luna, sage, felix, milo, atlas, and orion, we help creators select the tone that best fits their brand without relying on subjective hype.
Data-Driven Reviews
Our reviews rely on concrete metrics rather than opinion. We track generation speed, character limits, and language support to give you a realistic expectation of performance.
Transparent Pricing Analysis
We break down the $10 per million character cost and the $0.50 trial structure. This helps you budget accurately for long-form voiceovers or frequent short clips without hidden fees.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. We check for listener fatigue, ensuring the voice remains pleasant and clear during hour-long streams without becoming grating.
23 languages
Any voice speaks any of these languages.
Sources and further reading
- Wikipedia
- curl documentation
- Node.js documentation
- W3C WAI: making audio and video accessible
- LibriTTS-R paper (arXiv 2305.18802)
- LibriTTS-R corpus (OpenSLR 141)
- TRON developer hub: TRC-20
- Creative Commons Attribution 4.0
- LibriVox public-domain audiobooks
- Python documentation
- openai-node on GitHub
- openai-python on GitHub
