Educational Robotics With On-Device AI: An Interactive System Based on the Phi-3 Mini Small Language Model Running on Conventional Hardware

• Vitor Amadeu Souza UniFOA https://orcid.org/0009-0002-1857-6800

DOI: https://doi.org/10.5753/semish.2026.23148

Resumo

This work presents an educational robotics system integrating speech recognition, a Small Language Model (SLM), and speech synthesis on GPU-free hardware. It uses Google Speech Recognition, Phi-3 Mini via Ollama with 4-bit quantization, pyttsx3, and Pygame. Experiments on an Intel Core i5-3570 with 16 GB RAM evaluated five interactions. The average latency was 30.16 s, with speech recognition (6.80 s), SLM processing (8.06 s), and synthesis (15.30 s). Speech synthesis was the main bottleneck (50.73%), followed by SLM (26.72%) and recognition (22.55%). Results demonstrate feasibility on low-cost hardware while highlighting the need for latency optimization for more natural interaction.

Referências

J. Wang et al., “Large language models for robotics: Opportunities, challenges, and perspectives,” Journal of Automation and Intelligence, vol. 4, no. 1, pp. 52–64, Mar. 2025.

M. Abdin et al., “Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone,” Microsoft Research, 2024. Available at: [link].

D. Gouaillier et al., “The NAO humanoid: a combination of performance and affordability,” arXiv preprint arXiv:0807.3223, 2008.

A. C. R. Ribeiro, D. A. C. Barone and L. E. P. Mizusaki, “ROBO+ EDU: Project and Implementation of Educational Robotics in Brazilian Public Schools,” in Robot Intelligence Technology and Applications 3, Cham: Springer, 2015, pp. 495–503.

AI FOR GOOD, “The future of educational robotics: enhancing education, bridging the digital divide, and supporting diverse learners,” ITU AI for Good, Dec. 16, 2024. Available at: [link].

CRESTA, “Engineering for Real-Time Voice Agent Latency,” Cresta Blog, 2024. Available at: [link].

BENTOML, “Exploring the World of Open-Source Text-to-Speech Models,” Ben-toML Blog, 2024. Available at: [link].

X. Li et al., “Cm-tts: Enhancing real-time text-to-speech synthesis efficiency through weighted samplers and consistency models,” in Findings of the Association for Computational Linguistics: NAACL 2024, 2024, pp. 3777–3794.

A. Vaswani et al., “Attention Is All You Need,” in Advances in Neural Information Processing Systems 30 (NIPS 2017), 2017. Available at: [link].

A. Faria et al., “Toward zero oracle word error rate on the switchboard benchmark,” arXiv preprint arXiv:2206.06192, 2022.

MICROSOFT RESEARCH, “Microsoft researchers achieve new conversational speech recognition milestone,” Microsoft Research Blog, Jun. 13, 2017. Available at: [link].

A. Ferraro et al., “Benchmarking open source and paid services for speech to text: an analysis of quality and input variety,” Frontiers in Big Data, vol. 6, 2023.

T. Belpaeme et al., “Social robots for education: A review,” Science Robotics, vol. 3, no. 21, p. eaat5954, 2018.

S. Bögels and F. Torreira, “Listeners use intonational phrase boundaries to project turn ends in spoken interaction,” Journal of Phonetics, vol. 52, pp. 46–57, Sept. 2015.

S. C. Levinson and F. Torreira, “Timing in turn-taking and its implications for processing models of language,” Frontiers in Psychology, vol. 6, p. 731, 2015.

D. J. Souza and M. A. Silva, “CEMAO OS: A GNU/Linux distribution for obsolete computer labs,” Undergraduate Thesis — IF Baiano, 2019. Available at: [link].

A. Zhang, “SpeechRecognition,” PyPI, version 3.11, 2017. Available at: [link].

D. Amos, “The Ultimate Guide To Speech Recognition With Python,” Real Python, Sept. 7, 2023. Available at: [link].

N. M. Bhat, “pyttsx3: Text to Speech (TTS) library for Python 3,” PyPI, version 2.99, Jul. 8, 2025. Available at: [link].

GEEKSFORGEEKS, “Text to Speech by using pyttsx3 - Python,” Apr. 14, 2025. Available at: [link].

T. Stivers et al., “Universals and cultural variation in turn-taking in conversation,” PNAS, vol. 106, no. 26, pp. 10587–10592, 2009.

S. G. Roberts, F. Torreira and S. C. Levinson, “The effects of processing and sequence organization on the timing of turn taking: a corpus study,” Frontiers in Psychology, vol. 6, p. 509, 2015.

J. Nielsen, “Response Time Limits,” Nielsen Norman Group, 2024. Available at: [link].

H. F. Bush et al., “An Investigation of the Effect of Network Latency on Pedagogic Efficacy: A Comparison of Disciplines,” Contemporary Issues in Education Research, vol. 1, no. 4, pp. 11–26, 2008.

MOZILLA OCHO, “Llamafile 0.7 release notes: AVX-512 support,” GitHub, Apr. 1, 2024. Available at: [link].

LENOVO and INTEL, “AI Inferencing on Intel CPU-Powered Lenovo Servers: Accelerating LLM Performance with Intel Technology,” Lenovo Press, 2024. Available at: [link].