FluidAudio is an open-source Swift SDK from FluidInference that puts frontier audio AI models directly inside your iOS and macOS apps. Built on CoreML and the Apple Neural Engine, it delivers speech-to-text, text-to-speech, voice activity detection and speaker diarization entirely on-device — no servers, no per-minute API bills, no audio leaving the user’s phone or Mac. It has about 3,100 stars on GitHub and is updated continuously in 2026.
The SDK packages state-of-the-art open models — including NVIDIA’s Parakeet for fast, accurate automatic speech recognition and Kokoro-style text-to-speech — as optimized CoreML bundles with a clean Swift API. Streaming transcription, word-level timestamps and speaker identification/diarization are handled for you, and models download on demand so app size stays lean. Because inference runs on the Neural Engine through AVFoundation-friendly pipelines, latency stays low enough for live dictation, meeting capture and voice interfaces, and battery-friendly enough for mobile. FluidAudio is the open-source core that also powers FluidInference’s own products, so it benefits from production hardening: developers get the same components used in shipped apps, with documentation and sample projects to start from.
FluidAudio is completely free and open source under the Apache-2.0 license, including for commercial apps. There are no usage fees because all processing happens on the user’s device.
FluidAudio is for iOS and macOS developers who need fast, private, offline-capable speech recognition, synthesis or speaker analytics in their apps without depending on cloud speech APIs.