Child Speech Recognition in Human-Robot Interaction: Problem Solved?
Ruben Janssens, Eva Verhelst, Giulio Antonio Abbo, Qiaoqiao Ren, Maria Jose Pinto Bernal and Tony BelpaemeTL;DR
We revisit a paper on child speech recognition and using the same process, we test the performance of Whisper models. There is still room for improvement but they are perfectly usable. Read here a nice in-depth article on this work.
PUBLICATIONS IN THIS RESEARCH LINE
Collabs
Abstract
Automated Speech Recognition shows superhuman performance for adult English speech on a range of benchmarks, but disappoints when fed children’s speech. This has long sat in the way of child-robot interaction. Recent evolutions in data-driven speech recognition, including the availability of Transformer architectures and unprecedented volumes of training data, might mean a breakthrough for child speech recognition and social robot applications aimed at children. We revisit a study on child speech recognition from 2017 and show that indeed performance has increased, with newcomer OpenAI Whisper doing markedly better than leading commercial cloud services. Performance improves even more in highly structured interactions when priming models with specific phrases. While transcription is not perfect yet, the best model recognises 60.3% of sentences correctly barring small grammatical differences, with sub-second transcription time running on a local GPU, showing potential for usable autonomous child-robot speech interactions.
ACM Reference
Ruben Janssens, Eva Verhelst, Giulio Antonio Abbo, Qiaoqiao Ren, Maria Jose Pinto Bernal, and Tony Belpaeme. 2025. Child Speech Recognition in Human-Robot Interaction: Problem Solved? In Social Robotics (ICSR + AI '24), Springer Nature Singapore, Singapore, 476–486.BibTeX
@InProceedings{10.1007/978-981-96-3519-1_43,
author="Janssens, Ruben
and Verhelst, Eva
and Abbo, Giulio Antonio
and Ren, Qiaoqiao
and Bernal, Maria Jose Pinto
and Belpaeme, Tony",
editor="Palinko, Oskar
and Bodenhagen, Leon
and Cabibihan, John-John
and Fischer, Kerstin
and {\v{S}}abanovi{\'{c}}, Selma
and Winkle, Katie
and Behera, Laxmidhar
and Ge, Shuzhi Sam
and Chrysostomou, Dimitrios
and Jiang, Wanyue
and He, Hongsheng",
title="Child Speech Recognition in Human-Robot Interaction: Problem Solved?",
booktitle="Social Robotics",
year="2025",
publisher="Springer Nature Singapore",
address="Singapore",
pages="476--486",
abstract="Automated Speech Recognition shows superhuman performance for adult English speech on a range of benchmarks, but disappoints when fed children's speech. This has long sat in the way of child-robot interaction. Recent evolutions in data-driven speech recognition, including the availability of Transformer architectures and unprecedented volumes of training data, might mean a breakthrough for child speech recognition and social robot applications aimed at children. We revisit a study on child speech recognition from 2017 and show that indeed performance has increased, with newcomer OpenAI Whisper doing markedly better than leading commercial cloud services. Performance improves even more in highly structured interactions when priming models with specific phrases. While transcription is not perfect yet, the best model recognises 60.3{\%} of sentences correctly barring small grammatical differences, with sub-second transcription time running on a local GPU, showing potential for usable autonomous child-robot speech interactions.",
isbn="978-981-96-3519-1"
}