GoKawiilNew 'Whistle' model brings on-device speech-to-text in 16.9 MB
A new speech recognition model called Whistle has been released for mobiles, wearables, robots, smart home devices, automotive systems and microcontrollers. It ships as a single 16.9 MB file that runs entirely on-device via CPU with no dependencies, handling transcription, word-level timestamps and speech embeddings for seven languages without sending audio off the device.
GoKawiil's interpretation of the reporting above, not reported fact.
Running recognition fully on-device without cloud processing could appeal to privacy-conscious applications and low-power or offline hardware where connectivity is unreliable. Sharing its engine, container and quantisation approach with the Needle model suggests the developers are building a shared infrastructure for compact, edge-deployable AI tools rather than one-off products.
- Whistle is a 16.9 MB speech recognition model designed for edge devices like wearables and microcontrollers
- It performs transcription, word timestamps and speech embeddings entirely on-device, supporting seven languages
- It shares its underlying C++ engine, container and quantisation method with the existing Needle model
Raspberry Pi 4 Model B — If you're excited about tiny on-device speech models like Whistle, a Raspberry Pi is the perfect playground to test CPU-only inference without cloud dependencies. Its low power draw and compact size make it ideal for embedding offline voice transcription into smart home or robotics projects. Pair it with a USB microphone and you've got a self-contained speech-to-text node in minutes.
See Raspberry Pi 4 Model B on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: cactuscompute.com, 2026-10-08
Published there as: “Whistle: Speech to Text in 16.9 MB”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.