From Voice to Shell: A SLM-Based Assistant for IoT Maintenance Tasks on the Edge
Year: 2025
Type: article
Abstract: IoT deployment relies on complex command-line interfaces that handle multiple commands difficult to master, especially for non-expert or novel technicians and maintainers. Spoken dialogue arises as the most natural interaction for providing assistance to humans. Therefore, this study presents a privacy-preserving multilingual voice assistant, completely run on edge devices, that transforms natural language instructions into executable commands for heterogeneous IoT deployments. Two contexts are considered: docker commands, whose options and arguments are publicly accessible, and an emergency luminaires system commands, a proprietary deployment whose information is non previously available to pretrained models. Multiple datasets for both contexts have been drafted to fine-tune and evaluate proposed models. A benchmark and its results for comparing proposed models is exposed, as well as a functional implementation of the voice assistant, composed of models from the comparison. The proposed system achieves 3.15 WER and 98.57% semantic similarity, for transcribing voice data, and takes only 2.8s to generate a rightful command with 81.36% accuracy and not needing GPU acceleration, demonstrating reasonable performance on tested edge devices. This study findings show that carefully fine-tuned Small Language Models (SLMs) deliver high efficiency for speech-driven IoT devices management at a fraction of the computational cost, eliminating Cloud dependence and protecting sensitive audio data.
Source: IEEE Internet of Things Journal
Authors: Ángela Gijón Flores, Cristina Bolaños Peño, Henry Llumiguano Solano, Jesús Fernández-Bermejo Ruiz, Félix Jesús Villanueva Molina, Fernando Rincón Calle
Institution: University of Castilla-La Mancha
Language: English
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,645
Awards: G1175115578, G2262748287, G24120701, G2601397997, G294895050, G300979063, G3480869486, G3681454997, G4042783231, smart Mirrors supporting healthier lives of Adolescents and Young Adults after cancer, G425668148, G464105488, G5080475149, G5195188938, G528262178, G6218905885, G7084143925, G7127239378, G800104512, G8617047575, G866412001, G8888022036
OpenAlex ID: W7105612452