Fish Audio as Production Control Plane

The real prize in voice infrastructure is owning the live call loop, not just speaking the last sentence. If Fish Audio handles when the agent should talk, when it should stop, how calls are performing, and where failures happen, it stops being an interchangeable TTS vendor and becomes the system developers wire into production, monitor every day, and tune around.1 2 3 4 5

• Turn detection and interruption handling sit in the critical path of every call. LiveKit exposes them as core agent controls, because small timing errors create awkward pauses, talk overs, and dropped intent. A vendor that owns those controls shapes call quality more than the voice itself.2 3 4
• Observability creates stickiness because teams debug latency, transcripts, end of turn timing, and tool failures in one place. LiveKit and Retell both position monitoring and analytics as part of the operating stack, which shows where value is moving in voice agents.3 5 6
• Fish Audio already reaches the market through Retell, LiveKit, Telnyx, and Sanas, and Fish Audio was at $21M annualized revenue as of July 31, 2026. Retell was estimated at $60M annualized revenue in April 2026, which shows the layer above raw synthesis is already large and monetizable.1 7 8

The next step is a stack where speech generation, turn taking, call controls, and analytics are sold together as one runtime for production voice apps. If Fish Audio keeps moving upward into that control layer, it can capture more revenue per minute, become harder to replace, and compete where budgets are set by call outcomes rather than audio usage alone.1 3 5 6