Full speech pipeline in native Swift/MLX – ASR, TTS, speech-to-speech, on-device (github.com)

Apple has unveiled a comprehensive suite of AI speech models developed for its Silicon processors using Swift and MLX, enhancing on-device functionalities such as automatic speech recognition (ASR), text-to-speech (TTS), and speech-to-speech capabilities. Key models include Qwen3-ASR for speech-to-text transcription supporting 52 languages, high-quality Qwen3-TTS capable of synthesizing speech in multiple accents, and PersonaPlex, which allows for full-duplex speech interactions. Additional tools, such as voice activity detection (VAD) and speaker diarization, further enrich the ecosystem. This advancement is significant for the AI/ML community, as it makes sophisticated speech-processing features accessible directly on Apple devices without relying on cloud services, ensuring faster response times and enhanced privacy. The models are optimized for device performance, capitalizing on the capabilities of Apple’s Neural Engine. Developers can leverage this robust toolkit in applications, with various model sizes and configurations enabling tailored usage for specific needs. This positions Apple at the forefront of speech processing technology, critical for the future of conversational AI and smart applications.