# S1

Fish Audio's earlier voice model, kept available for existing integrations. S1 exposes an explicit emotional vocabulary rather than interpreting free-form direction, which makes its output more predictable to script.

- **Model ID:** `fish-audio/s1`
- **Type:** speech
- **Providers:** fish-audio
- **Context window:** Not applicable
- **Maximum output tokens:** 0
- **Pricing:** $15/1M input tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/s1

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { experimental_generateSpeech as generateSpeech } from 'ai';
import { gateway } from '@ai-sdk/gateway';
import { writeFile } from 'node:fs/promises';

const result = await generateSpeech({
  model: gateway.speechModel('fish-audio/s1'),
  text: 'Hello from the Vercel AI Gateway!',
  // Browse voices at https://fish.audio/app/discovery
  // Open a voice, then use "Copy Model Id" in its "..." menu.
  voice: '933563129e564b19a115bedd57b7406a',
});

await writeFile('speech.mp3', result.audio.uint8Array);
```

## About

S1 is Fish Audio's earlier text-to-speech model, a 4-billion-parameter model covering 13 languages including English, Chinese, Japanese, German, French, Spanish, Korean, Arabic, Russian, Dutch, Italian, Polish, and Portuguese. Fish Audio keeps it available for existing integrations.

The difference from the S2 generation is how you direct it. S1 uses an explicit vocabulary written in parentheses: more than 60 expressions spanning basic and advanced emotions, tone markers, and audio effects. Where the S2 models interpret free-form description, S1 matches against a defined set. That is more limiting and more predictable, which is the tradeoff to weigh.

On transcription-accuracy measures of its output, S1 posts a word error rate of 0.8 percent and a character error rate of 0.4 percent, and it ranked first on TTS-Arena2 at the time of its release.

Integration uses the AI SDK's speech generation function.

## What to consider

S1 is a previous generation. Fish Audio recommends S2.1 Pro for production, and language coverage is the clearest gap: 13 languages here against 83. For a new integration, start with S2.1 Pro unless the fixed emotional vocabulary is specifically what you want.

That vocabulary is the one reason to prefer S1. A defined set of expressions and tone markers produces more repeatable results than interpreted free-form direction, which matters when a script has to sound the same across regenerations.

Since it is maintained for existing integrations rather than actively developed, weigh how long you expect to depend on it before building something new on top.

## When to use

### Best For

- **Existing S1 integrations** already built against its parenthesis syntax
- **Repeatable delivery** from a fixed expression vocabulary rather than interpreted direction
- **Explicit audio effects** and tone markers selected from a defined set
- **The 13 supported languages** where wider coverage is not needed

### Consider Alternatives When

- **A new integration** is being built, where S2.1 Pro is the recommended production model
- **Wider language coverage** is needed, where S2.1 Pro supports 83 against 13
- **Free-form emotional direction** suits the workflow, where the S2 generation interprets plain language
- **Long-term support** matters, since this model is maintained rather than actively developed

## Best for

- **Existing S1 Integrations:** Already built against its parenthesis syntax
- **Repeatable Delivery:** A fixed vocabulary instead of interpreted direction
- **Explicit Audio Effects:** Tone markers chosen from a defined set
- **The 13 Supported Languages:** Where wider coverage is unnecessary

## Consider alternatives

- **New Integrations:** S2.1 Pro is the recommended production model
- **Wider Language Coverage:** S2.1 Pro supports 83 languages against 13
- **Free-Form Direction:** The S2 generation interprets plain language
- **Long-Term Support:** This model is maintained rather than actively developed

## Frequently asked questions

### Should I use S1 for a new project?

Usually not. Fish Audio recommends S2.1 Pro for production. Choose S1 only when its fixed emotional vocabulary is specifically what you need.

### How many languages does S1 support?

13, including English, Chinese, Japanese, German, French, Spanish, Korean, Arabic, Russian, Dutch, Italian, Polish, and Portuguese.

### How does emotional control work on S1?

Through an explicit vocabulary written in parentheses, covering more than 60 expressions across basic and advanced emotions, tone markers, and audio effects. The S2 generation instead interprets free-form description.

### Why choose a fixed vocabulary over free-form direction?

Repeatability. A defined set produces more consistent results across regenerations, which matters when a script has to sound the same each time it is produced.

### How accurate is S1 output?

On transcription-accuracy measures of its speech it posts a word error rate of 0.8 percent and a character error rate of 0.4 percent, and it ranked first on TTS-Arena2 at release.

### Does S1 support Zero Data Retention?

Zero Data Retention is not currently available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

## Links

- [Model page](https://vercel.com/ai-gateway/models/s1)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://docs.fish.audio/developer-guide/models-pricing/models-overview)
- [Provider pricing](https://docs.fish.audio/developer-guide/models-pricing/models-overview)