Jaco: An Offline Running Privacy-aware Voice Assistant
Daniel Bermuth, Alexander Poeppel, Wolfgang Reif
TL;DR
The paper addresses privacy concerns in voice assistants by proposing Jaco, an offline, multilingual assistant that can run on low-resource devices. It presents a modular, containerized architecture with a two-tier design (Satellites and Master) and an offline pipeline (wake word, STT, NLU, TTS) coordinated via MQTT, enabling safe on-device training and operation. Skills are easily added and isolated within containers, governed by a per-skill permission system and a store for sharing, inspection, and offline deployment. Benchmark results demonstrate competitive offline performance on standard SLU tasks, highlighting strengths in medium/low-noise conditions and in hardware-enabled scenarios, with noted challenges in pronunciation of artist names across languages. Overall, Jaco offers a privacy-preserving, extensible, and practical offline alternative to cloud-based voice assistants for real-world use on consumer hardware.
Abstract
With the recent advance in speech technology, smart voice assistants have been improved and are now used by many people. But often these assistants are running online as a cloud service and are not always known for a good protection of users' privacy. This paper presents the architecture of a novel voice assistant, called Jaco, with the following features: (a) It can run completely offline, even on low resource devices like a RaspberryPi. (b) Through a skill concept it can be easily extended. (c) The architectural focus is on protecting users' privacy, but without restricting capabilities for developers. (d) It supports multiple languages. (e) It is competitive with other voice assistant solutions. In this respect the assistant combines and extends the advantages of other approaches.