FluidInference/FluidAudio

FluidAudio is a Swift SDK for running audio AI models locally on Apple devices with inference offloaded to the Apple Neural Engine.

The tool addresses the need for low-latency, privacy-preserving audio processing on iOS and macOS by bundling state-of-the-art models for speech-to-text transcription, text-to-speech synthesis, voice activity detection, and speaker diarization. By offloading inference to the Apple Neural Engine rather than the CPU or GPU, the SDK minimizes memory consumption and power usage, making it suitable for background and always-on workloads. Models are provided as open-source implementations optimized for on-device deployment and can be integrated with minimal code.

Developers building audio-centric applications on Apple platforms should consider this tool if they need multiple audio AI capabilities without relying on cloud services. It suits projects requiring real-time processing, ambient computing features, or scenarios where privacy and latency are critical. The SDK is particularly valuable for applications that cannot tolerate network round-trips or want to avoid the overhead of GPU inference on mobile devices.

The project maintains active engagement with its user community through a dedicated Discord channel for feedback and feature requests. Development is oriented toward expanding model coverage, with plans to bring visual, language, and text-to-speech models to device. The team provides tooling for custom model conversion through a separate repository, enabling developers to optimize their own models for the platform.