Frequently asked questions
Is this the official GCP text to speech API?
No. We are an independent storefront for the EchoMouth TTS API. Our endpoint is compatible with the OpenAI speech endpoint structure, allowing you to use official OpenAI SDKs by simply changing the base URL to https://api.ao999.info/v1. We are not affiliated with OpenAI or any other brand.
Can I stream audio while it is generating?
No. The API returns the complete audio file in the response after synthesis finishes. There is no streaming, WebSocket, or real-time partial audio support. You must wait for the full file to be generated before receiving the data, which typically takes a few seconds for short phrases.
Do you support SSML for prosody control?
No. The API does not support SSML, phoneme dictionaries, or custom pronunciation lexicons. You control pacing and tone only by selecting one of the eight voice presets or adjusting the speed parameter between 0.25 and 4.0. Prosody is determined by the voice model and input text structure.
What happens if my API key is compromised?
You can issue a new key at any time from your account dashboard. When a new key is generated, the old one stops working immediately. Each account has one active key at a time, ensuring you can quickly revoke access if your credentials are exposed or misused.
Can I clone my own voice for a unique NPC?
No. We do not offer custom voice cloning, voice design, or the ability to upload your own voice recordings. You must choose from the eight available presets, which are based on licensed recordings from LibriTTS-R and LibriVox. Each voice has a distinct tone suitable for different NPC archetypes.
Does the API support emotion tags in the voice?
No. The voices do not support emotion tags or dynamic emotional shifts during generation. Each preset has a fixed character: for example, Nova is bright and upbeat, while Luna is soft and storytelling. You select the appropriate voice for the desired emotional tone of the dialogue line.
Is there a free tier for testing NPC dialogue?
Yes. Signing in with Google provides a $0.50 free trial credit, valid for 7 days, which covers about 50,000 characters. Accounts created with email and password do not receive this trial credit. You can test voice quality and integration before committing to USDT top-ups for production.
How do I handle text longer than 4096 characters?
Split your text into parts of up to 4096 characters and send separate requests for each part. Then, join the resulting audio files in your pipeline. This approach ensures each request stays within the character limit while allowing you to generate dialogue for longer NPC conversations or monologues.
What audio formats are supported for game engines?
The API supports mp3, wav, opus, flac, and pcm formats. These formats cover most game engine requirements, from compressed streaming (mp3, opus) to uncompressed editing (wav, flac) and raw data (pcm). Select the format that best fits your engine’s audio pipeline and storage constraints.
Do you store the generated audio files?
No. Generated audio is returned in the response and not stored by us. Input text is used only to produce the audio and is not used for training. We retain request metadata, such as time, character count, and status, solely for billing and usage tracking purposes.
Can I use the audio in commercial game projects?
Yes. The voices are based on licensed recordings from LibriTTS-R (CC BY 4.0) and LibriVox public-domain audiobook readings. You can use the generated synthetic speech in commercial projects. Each file carries an inaudible watermark to mark it as synthetic, but this does not restrict commercial usage.
Is there a rate limit for high-volume NPC generation?
Yes. The limit is 60 requests per minute per key, or 10 per minute before your first top-up. Requests are processed sequentially, so under load, a request may wait in a queue. For high-volume NPC dialogue, plan your generation batches to stay within these limits.
Do you offer SLA for production voice lines?
No. We do not offer an SLA, uptime guarantee, or enterprise contract. There is no on-premise install option or data-center choice. Support is provided via email only, with no phone number or live chat. This keeps costs predictable for freelancers and indie developers.
How is the synthetic watermark detected?
Every generated file carries an inaudible watermark that marks it as synthetic speech. This watermark is embedded in the audio data itself.
First request
curl
curl https://api.ao999.info/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.mp3Pipeline Integration
EchoMouth provides a GCP-compatible endpoint optimized for game developers needing deterministic NPC dialogue generation. The service mirrors the OpenAI speech endpoint structure, allowing you to use official OpenAI SDKs by simply changing the base_url to https://api.ao999.info/v1. This ensures seamless integration into existing automation pipelines without requiring custom client libraries.
OpenAI-Compatible Endpoint
Send text to POST https://api.ao999.info/v1/audio/speech and receive an audio file. The request shape matches the OpenAI standard, so your existing code works with a minor configuration change. This compatibility reduces boilerplate and keeps your asset production pipeline stable.
JSON Body Structure
The API accepts a JSON body with fields for model ("tts-1" or "tts-1-hd"), input (text up to 4096 characters), voice (one of 8 presets), response_format (mp3, wav, opus, flac, pcm), speed (0.25–4.0), and an optional language ISO code. Authentication is handled via an Authorization: Bearer header with your unique key.
Streaming Limitations
The API does not support streaming, WebSockets, or real-time partial audio. Requests are processed synchronously, returning the complete audio file upon completion. For long NPC scripts, split the text into parts of up to 4096 characters and join the resulting audio files in your build pipeline.
Voice Selection
Choose from eight deterministic voice presets based on licensed LibriTTS-R and LibriVox recordings. Each voice maintains consistent tonal characteristics across all generated NPC dialogue, ensuring uniform audio assets for your game worlds.
8 Preset Personas
Female presets include Aria (warm narrator), Nova (upbeat), Luna (storytelling), and Sage (instructional). Male options feature Felix (energetic promo), Milo (conversational), Atlas (neutral news), and Orion (deep documentary). All voices support any of the 23 available languages.
- 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
Gender and Tone Profiles
Select voices based on narrative role rather than strict gender. Atlas provides neutral delivery suitable for exposition, while Orion’s depth suits documentary-style lore. Felix’s energy drives active sequences.
Synthetic Watermarking
Every generated audio file contains an inaudible watermark marking it as synthetic speech. This allows easy identification of AI-generated dialogue assets within your production pipeline without affecting audio quality.
Language Support
EchoMouth supports 23 languages, allowing NPCs to speak in 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 deliver any supported language, ensuring consistent audio quality across diverse game worlds without requiring separate voice assets for each region.
23 Supported Languages
Our API covers the primary linguistic markets for global game releases. All eight voice presets (Aria, Nova, Luna, Sage, Felix, Milo, Atlas, Orion) support all 23 listed languages. This uniformity simplifies asset management for developers who need consistent voice characteristics across multiple localized versions of the same dialogue tree.
- Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish
Auto-Detection Logic
The API automatically detects language from the input script if you omit the language field. Latin script defaults to English, while other scripts are guessed from the input. This reduces payload size for simple English NPC interactions, though it may misidentify mixed-language dialogue.
Explicit Language Tags
For non-English Latin-script text (e.g., French or German), always set the language parameter to ensure accurate pronunciation. Without this tag, the engine defaults to English phonemes, resulting in unnatural speech. Explicit tags guarantee deterministic output for cross-lingual NPC interactions.
- Set
language: 'fr'for French text - Set
language: 'de'for German text - Omit field only for pure English input
Technical Constraints
Inputs are capped at 4,096 characters per request, including spaces and punctuation. This limit applies to each individual API call, requiring long NPC dialogue lines to be segmented before submission to the synthesis engine.
Audio formats are selected via the response_format field. Available outputs include mp3, wav, opus, flac, and pcm, allowing you to choose between compressed efficiency and lossless fidelity for your game assets.
4096 Character Limit
The API enforces a strict 4,096-character limit per request. This constraint applies to the input field and includes all spaces, punctuation, and special characters. Exceeding this limit results in an immediate rejection, ensuring deterministic processing for batched NPC dialogue generation.
Text Splitting Strategy
For dialogue exceeding the character cap, split the text into chunks of up to 4,096 characters. Generate each chunk separately and join the resulting audio files client-side. This approach maintains continuity while respecting the single-request boundary.
Note that requests are processed sequentially. Under load, longer texts may experience queue delays, so plan batch submissions accordingly to avoid pipeline bottlenecks during asset production.
Audio Format Options
Select the output format using the response_format parameter. Options include mp3, wav, opus, flac, and pcm. Choose based on your engine's requirements: mp3 or opus for storage efficiency, or wav/flac for lossless quality in post-production.
PCM is available for raw data pipelines. Ensure your implementation handles the specific binary or base64 encoding returned by the API for the chosen format.
- mp3: Compressed, widely supported
- wav: Uncompressed, standard for game engines
- opus: High efficiency, low bitrate
- flac: Lossless compression
- pcm: Raw linear audio data
Performance Metrics
Synthesis speed approximates real-time at roughly 25 characters of text per second for standard outputs. Short phrases require a few seconds to process, while longer dialogues generate at a steady pace proportional to length.
Real-Time Generation Ratio
Input text is processed sequentially. The system generates audio at a rate that generally matches playback speed, allowing developers to estimate generation time based on character count without complex calculations.
Queue Processing Behavior
Requests are processed one after another in a single queue. Under heavy load, additional requests may wait in line before synthesis begins, causing variable delays between submission and completion.
- Requests are processed sequentially.
Latency Expectations
Short sentences typically complete within a few seconds. Longer texts scale linearly with duration. During peak usage, latency increases due to queue waiting time rather than engine slowdown.

Pricing Model
Billing is calculated strictly per character, including spaces and punctuation. The rate is $0.01 per 1,000 characters, or $10 per million characters. This linear model ensures predictable costs for large NPC dialogue batches without hidden tier fees.
Per-Character Billing
Every character in your input text counts toward the total. Short phrases and long monologues are billed at the same uniform rate. This transparency allows precise budgeting for voice-over assets before generation begins.
Failed Request Refunds
If synthesis fails, the charge is returned in full. You only pay for successfully generated audio files. This reduces risk when testing edge cases or unusual character sets in dialogue scripts.
No Minimum Spend
There are no monthly plans or required minimums. Pay only for what you generate. Unused balance does not expire, allowing you to top up only when needed for upcoming production cycles.
Payment Infrastructure
EchoMouth operates on a direct payment model using USDT on the TRON network (TRC-20). We eliminate traditional payment friction by accepting only crypto transfers, ensuring transparent, borderless billing for freelance audio engineers and game developers without intermediary fees or currency conversion delays.
USDT TRC-20 Only
Payments are processed exclusively via USDT on the TRC-20 network. This single-currency approach simplifies accounting and ensures that your credits reflect the exact value transferred, without variance from banking holidays or credit card processing windows.
- USDT (TRC-20) only
- No credit cards or PayPal
- No VAT invoices or subscriptions
Instant Credit Top-up
Once your transfer is confirmed on the blockchain, your balance is updated automatically within minutes. There is no manual approval process, allowing you to resume batch NPC dialogue generation immediately after funding your account.
No Subscription Models
We do not enforce monthly plans or minimum spend requirements. You pay only for the characters you consume, and unused balance never expires. This structure aligns costs directly with your production volume, whether you are generating ten lines or ten thousand.
Developer Experience
Authentication relies on a single active API key per account, simplifying integration for freelance audio engineers managing multiple game projects. Issuing a new key immediately invalidates the previous one, ensuring no concurrent usage drift across environments.
One Active Key Policy
Maintain a clean security posture by rotating credentials without legacy overlap. The system ensures only the most recent key remains valid for POST requests to the speech endpoint.
Instant Key Revocation
Revocation is immediate. Once a new key is generated via the dashboard, the old key stops processing requests instantly, preventing unauthorized use during deployment transitions.
Usage Dashboard
The page provides visibility into your API key, current balance, and character consumption. Track usage metrics and top-up status directly from the /get-api-key/ page without external monitoring tools.
- View current API key and balance
- Monitor character consumption
- Access ready-to-use cURL examples
- Check trial credit status
Use Case: NPC Dialogue
NPC dialogue often requires generating hundreds of lines per scene. The API processes these in batches by splitting text into 4096-character chunks, synthesizing each segment, and joining the resulting audio files. This ensures deterministic output for consistent asset pipelines.
Batch Generation Workflow
Send a list of dialogue strings via the tts-1 or tts-1-hd model. The system returns individual audio files which your build script concatenates. Failed requests incur no cost, reducing risk during large batch runs.
- Split long texts into ≤4096 character segments
- Concatenate MP3/PCM outputs locally
- No per-request overhead beyond character count
Character Consistency
Assign a single voice preset (e.g., aria or atlas) to a specific NPC. Since voices are deterministic presets, the same text input always produces the same audio waveform. This eliminates variance between recording sessions or random generations.
- Fixed voice profiles per character
- Deterministic synthesis output
- Inaudible synthetic watermark in all files
Volume Scaling
Generation speed averages roughly real-time (25 characters per second). For large projects, queue processing may introduce slight delays under load. With a rate limit of 60 requests per minute, parallelize requests across multiple API keys if needed.
- Rate limit: 60 req/min per key
- Parallelize across keys for higher throughput
- No minimum spend or subscription tiers
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. Unlike our model, Azure requires a monthly Azure subscription and potentially higher minimum spend commitments.
23 languages
Any voice speaks any of these languages.
