Fish Audio S2 Pro tutorial and ElevenLabs v3 comparison

Fish Audio S2 Pro is the subject of this sponsored review and website tutorial. The speaker compares speech with and without inline tags, using examples of whispering, emphasis, pauses and emotional delivery. He generally prefers the tagged samples, though those judgments reflect his listening preferences.

The walkthrough covers account creation, voice cloning from an audio sample and optional vocal separation for noisy recordings. The speaker says cloning needs no fine-tuning and notes that the uploaded reference affects the resulting voice. In the demonstrated website interface, a paid plan is required to keep a cloned voice private. He also demonstrates automatic tag suggestions and voices from the library.

The ElevenLabs v3 comparison produces mixed preferences. The speaker favors S2's resemblance to the reference voice and its opening chuckle, but prefers ElevenLabs for the sadness in one sample. These examples do not establish a general quality ranking.

For local AI use, he describes downloadable weights and inference and fine-tuning code, with a non-commercial restriction on the open weights. He reports a requirement of about 24 GB VRAM and says he has not yet made local inference run smoothly. This is not a local installation tutorial. A Hugging Face Space provides another way to try the model; he also identifies Qwen 3 30B Omni as a dataset transcription requirement for fine-tuning.