Fish Audio: S1

Fish Audio: S1 is a multilingual text-to-speech model from Fish Audio. Fish Audio is the provider behind the model, and S1 sits in the speech category on api14. It is designed for voice applications that need broad emotional expression, and it uses parenthetical controls to guide speaking style across its supported languages. If you are building anything where the spoken delivery matters as much as the words, this is the kind of model the page is describing.

The model takes text as its only input and produces speech as its only output. There is no context window figure listed for it, and no supported request parameters are listed either. The details block records a maxOutput value of zero, which means there is no published output token ceiling to plan against on this page. Practically, you should treat S1 as a dedicated synthesis endpoint rather than a general chat model: you send text, you get audio back, and the expressive range comes from how you write the input text, not from tuning a set of sampling parameters.

That shape of model fits a fairly specific set of jobs. Narration and voiceover for video or audio content is the obvious one, since you can write the script and mark up delivery with parenthetical controls instead of re-recording takes. It also suits conversational agents and IVR style systems where you want the agent to sound engaged rather than flat, and accessibility work such as reading articles or documents aloud in more than one language. Any pipeline that generates speech at volume from text, and wants consistent emotional colour, is a reasonable candidate.

You call it through api14's OpenAI-compatible API using the model id "fish-audio/s1". Because the gateway follows the OpenAI request shape, you point your existing client at the api14 base URL and pass that id in the model field. Note that no supported parameters are listed for this model, so do not assume temperature, top_p, voice selection or similar knobs are accepted; send the text you want spoken and let the model handle synthesis. Check the live model page for the current parameter and endpoint details before wiring it into production.

On limits, the honest answer from the facts given is that this page does not state them. There is no context length, no maxOutput figure above zero, and no parameter list, so capacity and tuning behaviour are simply not documented here. If your workload depends on a known input size or on controlling output length precisely, verify those against the provider's own documentation first. For tasks that are not speech at all, a text or multimodal model on api14 will fit better, since S1 only accepts text and only returns speech.