Frequently asked questions
What is the base URL for the API?
The API base URL is https://api.2tiger.info/v1. All requests, including speech synthesis and voice listing, are directed to this endpoint. The standard speech endpoint is POST /v1/audio/speech. This URL structure ensures compatibility with existing client configurations designed for similar API standards.
Which SDKs are compatible?
The official OpenAI SDKs are compatible because the request shape matches the OpenAI speech endpoint. You can use these SDKs by changing the base_url to https://api.2tiger.info/v1 and providing your LinguaForge API key. No proprietary SDK is required for integration.
How do I set the base URL?
Configure your SDK or HTTP client to use https://api.2tiger.info/v1 as the base URL. For OpenAI SDKs, update the base_url parameter accordingly. Ensure your Authorization header includes your Bearer token. This directs all requests to the correct LinguaForge processing endpoint.
What are the valid voice IDs?
Valid voice IDs are: aria, nova, luna, sage, felix, milo, atlas, and orion. These eight presets cover various tones from warm narrators to deep documentaries. Each voice can speak any of the 23 supported languages, allowing flexible selection for your audio corpus needs.
Can I use custom models?
No custom models are available. The API offers two model options: tts-1 and tts-1-hd. Both use the same engine and pricing. You cannot upload custom voices or design new ones. The models provide consistent, watermark-marked synthetic speech suitable for machine learning training pipelines.
How do I specify the language?
Include the optional 'language' field with an ISO code in your request. If omitted, the system guesses the language based on the script, defaulting to English for Latin text. Explicitly setting the language ensures accurate synthesis for non-English Latin-script texts like German or Spanish.
What happens if the input is too long?
The input text is limited to 4096 characters per request. If your text exceeds this limit, you must split it into smaller parts. Each part is synthesized separately, and the resulting audio files are joined to form the complete output. This ensures consistent processing.
Do I need to split text manually?
Yes, you must split text exceeding 4096 characters manually before sending requests. The API does not handle automatic splitting or joining. You are responsible for dividing the text, sending multiple requests, and concatenating the returned audio files to create the final corpus segment.
How is speed controlled?
Control speed using the 'speed' parameter in your request body. The value must be a float between 0.25 and 4.0. This allows you to adjust the generation rate to fit your dataset requirements. The default speed produces audio at roughly real-time length.
What is the maximum speed value?
The maximum speed value is 4.0. The minimum is 0.25. Values outside this range will be rejected. Adjusting speed affects the duration of the generated audio. Faster speeds reduce generation time.
Is SSML supported in the input?
No, SSML is not supported. The API accepts plain text input only. There are no phoneme dictionaries or custom pronunciation lexicons available. You cannot use SSML tags to control pauses, emphasis, or pronunciation within the input string for your synthetic speech generation.
How do I handle rate limits?
The rate limit is 60 requests per minute per key. New accounts start with 10 requests per minute until the first top-up. Monitor your usage via the dashboard. Implement exponential backoff in your client to handle 429 errors gracefully during high-throughput corpus generation jobs.
What is the retry policy?
There is no SLA or uptime guarantee, so define your own retry policy. Failed requests are free, so retrying on failure incurs no cost. For transient errors, implement standard retry logic with delays. Requests are processed sequentially, so queueing may occur under load.
Can I stream the audio response?
No, streaming is not supported. The API returns the complete audio file in the response body. There is no WebSocket or real-time partial audio support. You must wait for the full file to be generated and received before processing it for your ML pipeline.
First request
curl
curl https://api.2tiger.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.mp3Python (openai)
from openai import OpenAI
client = OpenAI(base_url="https://api.2tiger.info/v1", api_key="tts-...")
with client.audio.speech.with_streaming_response.create(
model="tts-1", voice="nova", input="Hello from Python.", response_format="mp3"
) as response:
response.stream_to_file("hello.mp3")Node.js (openai)
import OpenAI from "openai";
import fs from "node:fs";
const client = new OpenAI({ baseURL: "https://api.2tiger.info/v1", apiKey: process.env.TTS_API_KEY });
const res = await client.audio.speech.create({ model: "tts-1", voice: "milo", input: "Hello from Node.js." });
fs.writeFileSync("hello.mp3", Buffer.from(await res.arrayBuffer()));| voice | |
|---|---|
aria | female |
nova | female |
luna | female |
sage | female |
felix | male |
milo | male |
atlas | male |
orion | male |
| language | |
|---|---|
ar | العربية |
zh | 中文 |
da | Dansk |
nl | Nederlands |
en | English |
fi | Suomi |
fr | Français |
de | Deutsch |
el | Ελληνικά |
he | עברית |
hi | हिन्दी |
it | Italiano |
ja | 日本語 |
ko | 한국어 |
ms | Bahasa Melayu |
no | Norsk |
pl | Polski |
pt | Português |
ru | Русский |
es | Español |
sw | Kiswahili |
sv | Svenska |
tr | Türkçe |
Authentication and Keys
LinguaForge authenticates requests via a Bearer token passed in the Authorization header. Each account maintains a single active API key. Send Authorization: Bearer tts-<your_key> with every request to the https://api.2tiger.info/v1 base URL. Invalid or expired tokens return a 401 Unauthorized error.
Keys are issued upon account creation. You can rotate keys at any time via the dashboard. When a new key is generated, the previous key becomes invalid immediately. Ensure your integration handles this atomic switch to avoid service interruption during maintenance.
Bearer Token Usage
Include the header Authorization: Bearer <key> in all POST requests to /v1/audio/speech. The token must be sent as a string without the Bearer prefix in the value itself if your SDK handles it automatically, otherwise append it explicitly. Use HTTPS exclusively.
- Format:
Authorization: Bearer tts-xxxxxxxx - Header name:
Authorization - Protocol: HTTPS required
Key Rotation Policy
Generate a new key from the /get-api-key/ dashboard at any time. The transition is instant. Old keys stop working immediately upon generation of the new one. Update your environment variables or configuration files promptly to prevent request failures during deployment cycles.
There is no grace period for old keys. Plan rotations during low-traffic windows if your pipeline is critical, though the atomic swap minimizes overlap issues.
- Instant invalidation of old keys
- No grace period
- Generate via dashboard
Single Active Key Rule
Each account holds exactly one active key. Issuing a new key invalidates the current one. This simplifies security audits and key management for indie developers. You do not need to manage multiple key versions simultaneously.
If you need a separate key for development, you must generate a new one, which will revoke your production key. Use the same key across all environments if possible to avoid downtime.
- One key per account
- Old key revoked instantly
- No concurrent keys
Base URL Configuration
Configure your API client by pointing the base URL to https://api.2tiger.info/v1. This structure ensures compatibility with standard HTTP clients and allows you to use existing OpenAI SDKs by simply overriding the default endpoint. The base path remains identical to the official OpenAI speech endpoint, requiring only a change in the host address.
OpenAI SDK Compatibility
The API accepts the same request shape as OpenAI’s POST /v1/audio/speech. By setting your SDK’s base_url to https://api.2tiger.info/v1, you can utilize the official OpenAI libraries without rewriting your integration logic. This ensures consistent parameter handling for model, voice, and format selections.
Endpoint Modification
To direct requests to LinguaForge, update your client configuration to use https://api.2tiger.info/v1 as the base URL. All subsequent calls, such as /audio/speech or /voices, append directly to this base. This single configuration change routes all traffic through the correct gateway without requiring additional path prefixes.
- Base URL:
https://api.2tiger.info/v1 - Speech Endpoint:
/audio/speech - Voices Endpoint:
/voices
SDK Version Requirements
Any HTTP client or SDK supporting standard JSON payloads and Bearer token authentication is sufficient. No proprietary SDK is required. Ensure your library supports dynamic base URL configuration, as most modern HTTP clients allow this via a simple configuration object or environment variable. The interface relies on standard REST conventions.
Speech Generation Endpoint
The core function of the API is audio synthesis via a single HTTP endpoint. Send a POST request to https://api.2tiger.info/v1/audio/speech with a JSON body containing the text, voice model, and desired format. The service returns the generated audio file directly in the response stream.
This endpoint supports the standard OpenAI request shape, allowing immediate integration with existing SDKs by adjusting the base URL. Each request processes one batch of text, converting it into the specified audio format for immediate download or storage in your ML pipeline.
POST /v1/audio/speech
Send a POST request to the speech endpoint. The request must include a valid Bearer token in the Authorization header. The response contains the audio data in the requested format, ready for immediate use in training datasets.
- Endpoint:
POST https://api.2tiger.info/v1/audio/speech - Method: HTTP POST
- Response: Audio file stream
JSON Body Structure
The request body is a JSON object. Required fields include model ("tts-1" or "tts-1-hd"), input (the text string), and voice (one of the eight presets). Optional fields include response_format, speed, and language.
The model field determines the engine quality, though both models use the same pricing tier. The voice field selects from the eight available presets, such as aria or atlas. The input field accepts the raw text to be synthesized.
model: "tts-1" or "tts-1-hd"input: String up to 4096 charactersvoice: Preset name (e.g., "nova")response_format: mp3, wav, opus, flac, or pcmspeed: 0.25–4.0language: ISO code (optional)
Input Text Limits
Each request supports a maximum of 4096 characters in the input field. This limit applies to the entire text string, including spaces and punctuation. For longer texts, split the content into chunks of 4096 characters or less.
Generate audio for each chunk separately and join the resulting files. This ensures reliable synthesis within the character limit. The API processes requests sequentially, so queueing may occur under load, but each chunk is handled independently.
- Max 4096 characters per request
- Includes spaces and punctuation
- Split longer texts manually
- Join output files after generation
Voice and Model Selection
The API supports two model variants, tts-1 and tts-1-hd, which utilize the same underlying engine and pricing structure. Both models accept identical input parameters and voice presets, allowing for straightforward switching without code refactoring. The distinction lies solely in the model name; both models use the same engine and pricing.
tts-1 Model Details
The tts-1 model delivers standard synthetic speech suitable for general machine learning datasets. It processes text efficiently, maintaining real-time generation speeds for long-form content. This model is optimal for high-volume labeling where minor audio artifacts do not impact model training accuracy.
tts-1-hd Model Details
The tts-1-hd model enhances audio clarity and naturalness, making it preferable for training data requiring precise phonetic representation. While it incurs the same per-character rate as tts-1, the improved fidelity can reduce manual correction efforts in downstream NLP pipelines.
Available Voice Presets
Select from eight distinct voice presets based on licensed recordings. Each voice can speak any of the 23 supported languages, ensuring consistent tonal characteristics across multilingual corpora. Voices are categorized by gender and tonal quality, such as the warm narrator aria or the deep documentary voice orion.
- Female: aria (warm), nova (bright), luna (soft), sage (calm)
- Male: felix (energetic), milo (friendly), atlas (neutral), orion (deep)
Response Formats
Specify the response_format parameter in your request to define the output audio container. Supported formats include MP3, WAV, FLAC, and PCM, allowing you to select the optimal balance between file size and fidelity for your ingestion pipeline.
MP3 Output
MP3 is the default format, providing a compressed audio stream that reduces bandwidth usage without significant quality loss for most text-to-speech applications.
WAV and FLAC Options
WAV and FLAC provide uncompressed, lossless audio data. Use these formats when preserving exact waveform integrity is critical for subsequent machine learning training steps or when your inference pipeline requires raw PCM data without decoding overhead.
- WAV: Uncompressed, widely supported.
- FLAC: Lossless compression, smaller file size than WAV.
PCM Raw Data
PCM returns raw audio data. This format is ideal for direct ingestion into audio processing libraries that handle their own encoding or when minimal latency is required.
- 16-bit depth.
- Little-endian byte order.

Speed and Language Parameters
The API supports a speed multiplier range of 0.25 to 4.0, allowing precise control over generation duration. Adjusting speed affects character-per-second output but does not alter voice characteristics or quality.
Speed Multiplier Range
Set the speed parameter to accelerate or decelerate speech. Values outside 0.25–4.0 are rejected. This is useful for synchronizing audio with video timelines or reducing processing time for short phrases.
- Default speed is 1.0.
- Lower values produce slower, clearer speech.
- Higher values increase throughput for large corpora.
Language Code Specification
Specify the ISO 639-1 language code (e.g., en, de, ja) in the request body. While most voices support all 23 languages, explicit codes ensure correct pronunciation and intonation, especially for Latin-script text that is not English.
- Supported: Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish.
- Not supported: Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, Bengali.
Auto-Detection Logic
If the language field is omitted, the system guesses the language from the script. Latin-script text defaults to English. For non-English Latin text, always set the language parameter to avoid incorrect pronunciation.
Error Handling and Status Codes
The API returns standard HTTP status codes to indicate request outcomes. A 200 OK response contains the generated audio file in the requested format. Non-successful codes indicate specific failure conditions such as invalid input or authentication errors.
Common HTTP Errors
Use 400 Bad Request for malformed JSON or invalid voice names. Use 401 Unauthorized if the Bearer token is missing or invalid. Use 404 Not Found for incorrect endpoints. Use 500 Internal Server Error for transient synthesis failures, which are refunded in full.
- 400: Invalid JSON structure
- 401: Missing or invalid API key
- 404: Endpoint not found
- 500: Synthesis failure (refunded)
Rate Limit Responses
Exceeding the 60 requests per minute limit returns a 429 Too Many Requests status. The response includes a Retry-After header indicating seconds to wait. Requests during the initial 10 RPM phase may also trigger this if exceeded.
- 429: Rate limit exceeded
- Retry-After header provided
- Reset every 60 seconds
Retry Logic
Implement exponential backoff for 429 and 500 errors. Do not retry 400 errors as they indicate permanent input issues. Wait for the duration specified in the Retry-After header before resuming requests to avoid queue congestion.
- Exponential backoff recommended
- Wait for Retry-After header
- Retry 500 errors once
Rate Limiting Rules
LinguaForge enforces strict rate limits to ensure stable throughput for high-volume corpus generation. Understanding these constraints is critical for maintaining consistent data ingestion pipelines without unexpected throttling delays.
The system applies a two-tier limit structure based on account funding status. Proper configuration of your client-side retry logic ensures efficient use of your allocated requests per minute.
60 Requests Per Minute
Standard accounts are permitted a maximum of 60 requests per minute per API key. This limit applies to the total number of POST requests to the speech endpoint, regardless of text length or voice selected.
Requests are counted within a sliding window. Exceeding this threshold results in a 429 Too Many Requests response, requiring your application to pause and retry after the window resets.
- Limit: 60 requests/minute per key
- Scope: All POST /v1/audio/speech calls
- Reset: Sliding window
Initial Limit Phase
New accounts or those with zero balance begin with a reduced limit of 10 requests per minute. This lower threshold persists until the first successful top-up is processed.
Once a minimum top-up of $5 USDT (TRC-20) is confirmed, the rate limit automatically increases to the standard 60 requests per minute. This applies to all active keys on the account.
- Default limit: 10 requests/minute
- Condition: Zero balance or new account
- Upgrade: Automatic after $5 top-up
Queue Behavior
Requests are processed one after another. If the queue is full, new requests may experience slight delays before processing begins.
Typical generation speed is roughly real-time (25 characters per second). Under heavy load, requests wait in the queue, potentially increasing total processing time for long texts.
Failed requests do not consume characters from your balance, but they do count toward the rate limit. Ensure your client handles 429 errors gracefully to avoid unnecessary retries.
- Processing: Sequential
- Delay: Possible under load
- Failed requests: Count toward limit
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. Balance key issuance with throughput needs, as each key has a fixed rate limit.
23 languages
Any voice speaks any of these languages.
