15,000 Natural-language Voice Directions

This feature turns voice generation from a single knob into line by line direction. Instead of making an entire script faster, slower, higher, or lower, Fish Audio lets the user tell the model how to perform each phrase, like whispering one clause, laughing on the next, then switching to a broadcast read. That matters because most real voice work is not one mood for 30 seconds, it is many small performance changes stitched together.1 2 3 4

• Fish Audio is built around inline control. Its API and model docs describe emotion and paralinguistic cue support inside the text itself, which fits the product workflow of pasting a script and directing specific words or sentences, not just setting one global voice style at the top.2 3
• The competitive contrast is with voice infrastructure that sells realism and low latency first. Cartesia centers Sonic for text to speech inside real time agents, while ElevenLabs has scaled around high quality synthetic voices and a broad audio toolset. Fish Audio is pushing deeper into direction following, which is closer to giving users a voice director inside the editor.5 6 7
• This also compounds the value of Fish Audio's voice library and cloning. A marketplace with 2M community voices and short sample cloning becomes more useful when each voice can be steered mid sentence across 80 plus languages, because the customer is not just picking a timbre, they are choosing a performer that can take instructions.1 2 4

The next battleground in voice AI is moving from sounding human to taking direction like a human. If Fish Audio keeps improving instruction following while expanding multilingual cloning and its voice marketplace, it can occupy the part of the stack where creators and developers decide not just whose voice to use, but exactly how every line should land.1 2 4 5