Fish Audio
Use Fish Audio's speech API directly with an API key from Fish Audio.
- Open Settings → Admin → Providers → Add provider.
- Choose Fish Audio and enter your API key.
- Choose a model. New setups use S2.1 Pro Free (
s2.1-pro-free); S2.1 Pro, S2 Pro and Drama 3 are paid. - Save, then use Check to verify access and generate a short worker sample.
- In Model & Voice, choose the provider and a voice. New setups use Sarah.
Whether the free model is available depends on your account and Fish's current service. OpenReader never retries a failed free-model request with a paid model.
Voices
All four models offer Sarah, Energetic Male, Adrian, Delia and Sheila. These fixed voice references keep narration consistent across passages. Enter any other Fish voice model ID under Custom voice ID.
The default option, at the bottom of the list, lets Fish choose a voice for each request. Existing setups that saved default keep it until a named voice is chosen and saved (or applied in the reader).
Users can also create their own voice from a short recording. Fish needs the exact words spoken in the sample, so the transcript is required. The sample is sent with each narration request; creating a voice does not use Fish's model-creation allowance.
Instructions and speed
Voice instructions accepts a short direction such as Warm, calm narration.
OpenReader adds it as a bracketed cue to each request after text filtering,
without changing the displayed passage or the word timing. Changing the
direction generates a separate, reusable version of the audio.
Native generation speed supports 0.5×–2×. Audio uses the usual worker alignment, playback cache and export paths.
Checking the provider
Check verifies API access with Fish's voice metadata endpoint. After saving, it also generates a short sample through the worker, since metadata access alone does not prove generation credit or model access.
API contract: Fish Audio text to speech. To use Fish models through OpenRouter's billing instead, see OpenRouter.