Frequently asked questions
Do I need the latest version of the OpenAI SDK?
No. The API follows the same request shape as the OpenAI speech endpoint, so any version of the OpenAI SDK works. You only need to change the base_url to https://api.98wi.info/v1 and use your PressSynth key. This allows you to use the official SDKs without waiting for a custom release or specific version requirements.
How do I handle long articles that exceed 4096 characters?
You must split the input text into parts of up to 4096 characters each. Send each part as a separate request to the POST /v1/audio/speech endpoint. Then, join the resulting audio files together in your application logic. This ensures every request stays within the character limit while preserving the full content of your article.
What is the default voice if I don't specify one?
The API does not have a single default voice preset. You must explicitly specify one of the eight available voices (e.g., aria, nova, luna, sage, felix, milo, atlas, orion).
Can I change the speed of the generated audio?
Yes. You can adjust the speed parameter in your request body to a value between 0.25 and 4.0. This allows you to speed up or slow down the narration to fit your editorial style. The default speed is 1.0.
Does the API support SSML tags in Python?
No. The API does not support SSML tags. You send plain text in the input field. If you need specific pauses or emphasis, you must manage this through your application logic or by adjusting the speed parameter. No phoneme dictionaries or custom pronunciation lexicons are available for fine-grained control.
How do I stream the audio directly in Python?
The API does not support streaming or WebSocket connections. It returns the complete audio file in the response body. You must wait for the full synthesis to complete before receiving the data.
What happens if I exceed the rate limit?
The limit is 60 requests per minute per key. If you exceed this, requests may be queued or rejected. Before your first top-up, the limit is 10 requests per minute. Failed requests are free, but excessive rejections may delay your workflow. Monitor your usage via the dashboard to stay within the allowed threshold.
Is the API key case-sensitive?
Yes, API keys are typically case-sensitive strings. You should use the exact key provided in your dashboard under the Authorization: Bearer header. If you issue a new key, the old one stops immediately. Ensure you copy the key exactly as displayed to avoid authentication errors when sending requests.
Can I use the same key for multiple projects?
Yes. You have one active key per account that can be used across multiple projects or applications. However, all projects share the same rate limit of 60 requests per minute. If you need higher throughput, consider managing multiple keys or scaling your request volume within the single key's limit.
How do I detect errors in the JSON response?
The API returns standard HTTP status codes and a JSON body on errors. Check the status code (e.g., 4xx for client errors, 5xx for server errors) and parse the JSON response for error details. Failed synthesis requests are free, so you can retry or adjust parameters based on the error message provided in the response body.
Do I need to install any other libraries for audio processing?
No. The API returns standard audio formats (mp3, wav, opus, flac, pcm). You only need a standard HTTP client library like requests or the OpenAI SDK. You do not need specialized audio processing libraries to receive the file. You can save the response content directly to a file or process it using your preferred audio library.
Is the base URL permanent or does it change?
The base URL https://api.98wi.info/v1 is permanent for this site. It does not change unless the site infrastructure is updated. You should hardcode this URL in your application configuration. The endpoint structure remains consistent, allowing you to rely on this address for all your text-to-speech integration needs.
How do I verify my API key is active?
You can verify your key by checking your account dashboard, which displays your current key, balance, and usage. You can also make a test request to the API. If the key is invalid or inactive, the API will return an authentication error. A new key issued replaces the old one immediately.
Can I use async Python with this API?
Yes. Since the API is a standard HTTP REST endpoint, you can use any async HTTP client like aiohttp or httpx with Python. The OpenAI Python SDK also supports async operations. You can send requests concurrently, but remember that the backend processes requests sequentially, so you may still experience queuing under high load.
First request
curl
curl https://api.98wi.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.98wi.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.98wi.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 setup
Access your unique API key via the /get-api-key/ dashboard page. Each account holds one active key, which you can regenerate at any time to secure your endpoint.
Include the key in your HTTP headers as a Bearer token: Authorization: Bearer tts-.... This token authenticates all requests to the API base URL.
Generate API key from dashboard
Sign in to your account on 98wi.info and navigate to the dashboard. Your active key is displayed alongside your current balance and usage statistics.
Bearer token format
Pass your key in the Authorization header using the Bearer scheme. The value must start with tts- followed by your unique identifier.
- Format:
Authorization: Bearer tts-12345abcdef - Case-sensitive; ensure no extra spaces.
Key rotation process
Generate a new key in the dashboard at any time. The previous key stops working immediately upon creation of the new one.
SDK installation and configuration
To generate speech using the PressSynth API, you can use the official OpenAI Python SDK, as the endpoint follows the same request shape. This allows you to leverage existing OpenAI client libraries by simply pointing them to our API base URL.
The integration requires minimal configuration. You install the standard package and override the base URL to direct requests to our infrastructure. This approach ensures compatibility with standard Python workflows used by publishers for automating article-to-audio pipelines.
Install OpenAI Python package
Install the latest version of the OpenAI Python package using pip. This provides the necessary client classes to interact with the speech endpoint without needing a proprietary SDK.
``bash
pip install openai
``
Set base_url variable
Configure the client to point to the PressSynth API base URL. Replace the default OpenAI endpoint with https://api.98wi.info/v1 so that all speech generation requests are routed correctly.
```python from openai import OpenAI client = OpenAI( base_url="https://api.98wi.info/v1", api_key="tts-your-api-key" ) ```
Configure environment variables
Store your API key securely in your environment variables rather than hardcoding it. Pass the key directly to the client constructor or use os.environ for cleaner deployment in production environments.
```python import os client = OpenAI( base_url="https://api.98wi.info/v1", api_key=os.environ["PRESS_SYNTH_API_KEY"] ) ```
- Key format:
tts-... - One active key per account
- Rotate keys anytime via dashboard
Making your first request
Send a POST request to the speech endpoint with your API key in the Authorization header. The API processes the text and returns the audio file directly in the response body.
This example uses the OpenAI Python SDK, configured to point to our base URL. It demonstrates a simple call to generate audio from a short news snippet.
Python code example
Install the official OpenAI package via pip. Set the base URL to https://api.98wi.info/v1 and provide your Bearer token. The code below initializes the client and calls the speech method.
JSON body structure
The request body must include the model (tts-1 or tts-1-hd), the input text, and the desired voice. You may also specify response_format and speed. All fields are standard for the speech endpoint.
- model: "tts-1" or "tts-1-hd"
- input: your text string
- voice: one of 8 presets
- response_format: mp3, wav, opus, flac, or pcm
Handling the response
The API returns the audio file directly. Save the binary content to a file with the correct extension matching your chosen format. If the synthesis fails, you receive a JSON error object with details.
Voice selection parameters
Retrieve the complete list of available voices and their supported languages by calling the GET /v1/voices endpoint. This returns metadata for all eight presets, allowing you to programmatically match voice styles to your article’s tone without hardcoding values.
List available voices
Use the /v1/voices endpoint to fetch all available presets. Each voice object includes its ID, gender, and supported language codes (all 23 languages are supported by all voices). This dynamic list ensures your application stays compatible with any future updates to the voice catalog.
Select by name
Pass the voice ID directly into your API request’s voice field. You can choose from eight distinct presets: Aria, Nova, Luna, and Sage for female tones; Felix, Milo, Atlas, and Orion for male tones. The system accepts the exact string name as defined in the API response.
- Aria: warm, clear narrator
- Nova: bright, upbeat
- Luna: soft, storytelling
- Sage: calm, instructional
- Felix: energetic, promo
- Milo: friendly, conversational
Voice characteristics
Each voice is derived from licensed recordings in the LibriTTS-R dataset, providing consistent broadcast-quality audio. All voices can speak all 23 supported languages. For instance, Atlas provides a neutral news tone, while Orion offers a deep documentary style. Select based on whether your content is conversational, instructional, or narrative-driven.
Input text limits
Each API call accepts a maximum of 4,096 characters for the input text. This limit applies to the raw string, including spaces and punctuation, rather than individual words or tokens.
4096 character limit per request
The API processes text in fixed chunks. If your input exceeds 4,096 characters, you must split the text into parts and join the resulting audio files. You must split longer content manually before sending it to the endpoint.
Splitting long articles
For full news articles or scripts, split the text into segments under the character limit. Generate audio for each segment separately, then join the resulting files to create a continuous broadcast-ready track.
- Split at logical sentence breaks to ensure smooth audio transitions.
- Process segments sequentially or in parallel depending on your rate limits.
- Concatenate output files using standard audio libraries or tools.
Character counting rules
Spaces and punctuation marks count toward the 4,096-character limit. Ensure your input string is trimmed and validated before sending to avoid silent truncation or errors.

Output formats
The API returns audio files in your chosen format via the response_format parameter. Select the format that best matches your delivery pipeline, whether for immediate web playback or archival storage.
MP3 for web streaming
MP3 is a standard format for web delivery. It offers a balance between file size and quality, making it suitable for embedding audio players in newsletters or publishing directly to your website.
WAV for high fidelity
WAV provides uncompressed audio. Use this format when you need lossless input for downstream systems.
FLAC and PCM options
FLAC offers lossless compression for smaller file sizes without quality loss. PCM provides raw, uncompressed data, suitable for specific processing pipelines that handle audio decoding themselves.
Rate limits and errors
The API enforces a strict rate limit of 60 requests per minute per key. Under heavy load, requests are processed sequentially, meaning longer texts may wait in a queue. Failed requests are free, ensuring you only pay for successful synthesis.
Before your first top-up, the limit is reduced to 10 requests per minute. This constraint applies to all keys and is enforced server-side. Exceeding the limit results in an HTTP 429 status code.
60 requests per minute limit
Once funded, your account supports 60 requests per minute. This allows for moderate batch processing of news articles or newsletters. Requests are queued if the limit is reached, preserving your balance but increasing latency.
- Limit resets every 60 seconds.
- Applies per API key.
- Queued requests wait for availability.
10 requests before top-up
New accounts start with a 10 requests per minute limit. This allows you to verify the API's behavior without immediate high-volume usage. The limit lifts automatically after your first USDT top-up is credited.
- Applies to trial and new keys.
- Lifts after first top-up.
- Applies to all keys.
Handling HTTP 429 errors
When you exceed the limit, the API returns HTTP 429 Too Many Requests. Implement exponential backoff in your Python client to retry automatically. Do not retry immediately, as the window has not yet reset.
Failed synthesis returns a different error code and is free. Only exceeded rate limits cause 429 errors. Check your dashboard for real-time usage stats.
- Retry with exponential backoff.
- Check dashboard for usage.
- Failed requests are free.
Speed and language controls
The API supports dynamic speed adjustment from 0.25 to 4.0, allowing publishers to adapt audio pacing for different content types without re-generating files. This flexibility is particularly useful for condensing lengthy newsletters or slowing down dense news scripts for clarity. Adjusting speed does not affect the character count or pricing, ensuring predictable costs regardless of playback rate.
Language handling is critical for global publications. While the system can guess language from the text script, explicit specification ensures accuracy, especially for mixed-language content or non-English Latin script text. This prevents mispronunciation and maintains broadcast quality across diverse audiences.
Adjustable speed 0.25–4.0
Set the speed parameter in your request to control playback rate. Values range from 0.25 (slow) to 4.0 (fast). This works seamlessly with all voice presets, allowing you to fine-tune the audio for specific editorial needs without additional generation steps.
- Speed affects playback duration, not character billing.
- Ideal for shortening long newsletters or clarifying dense text.
- Works with all 8 voice presets.
Setting ISO language codes
Pass the language parameter with a valid ISO code (e.g., en, de, ja) to enforce specific pronunciation. This is essential for non-English content or when the default detection might misinterpret the script. 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.
Note that Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, Bengali, and others are not supported. If your text is in an unsupported language, the API will attempt to process it, but pronunciation may be inaccurate.
- Supported: Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish.
- Unsupported: Vietnamese, Indonesian, Thai, Ukrainian, Romanian, Czech, Bengali.
- Explicit codes ensure consistent pronunciation across batches.
Auto-detection fallback
If the language field is omitted, the API attempts to detect the language from the input text script. Latin script text defaults to English. For non-English Latin-script text, such as German or Dutch, always set the language parameter to avoid incorrect English pronunciation.
This fallback mechanism simplifies integration for English-first publishers but requires explicit configuration for multilingual content to maintain broadcast quality.
- Latin text defaults to English if
languageis unset. - Set
languageexplicitly for non-English Latin scripts. - Auto-detection may misinterpret mixed-language content.
Listen to the 8 voices
Each sample below was generated by this API from the same sentence. The service provides eight preset voices based on licensed recordings from LibriTTS-R and LibriVox.
23 languages
Any voice speaks any of these languages.
Sources and further reading
- Wikipedia
- Creative Commons Attribution 4.0
- Node.js documentation
- LibriVox public-domain audiobooks
- Python documentation
- W3C WAI: making audio and video accessible
- LibriTTS-R paper (arXiv 2305.18802)
- Tether (USDT)
- TRON developer hub: TRC-20
- ISO 639 language codes
- MDN: Web Speech API
- LibriTTS-R corpus (OpenSLR 141)
