Scaling the Identification and Remediation of Reading Difficulties: A Gamified, Dynamic Automatic Speech Recognition Approach
Engineering, IT, Mathematics and Statistics
Please note; this internship is able to cover project costs for both domestic and international students.
ABOUT THE INDUSTRY PARTNER
Cinglevue International develops innovative solutions for organisations operating within the Early Years Learning to Higher Education space. Their goal is to make learning tangible and realisable for all and they have developed their Virtuoso enterprise learning platform to actively support teachers, learners, and the education community in this endeavour. Cinglevue actively seeks collaborative research opportunities in order to facilitate the transformation of transferred knowledge into technology that drives future features and products and builds credibility and authenticity based on common research interests. Cinglevue actively engage in research both internally and in collaboration with leading Australian universities to continually enhance the capabilities of their products in light of on-going advances in research, technology, and educational practice.
WHAT’S IN IT FOR YOU?
- Opportunity to collaborate with a multidisciplinary team (e.g., learning more about work conducted in the educational technology space and how this can advance your current research projects)
- Opportunity to gain/advance industry experience
- Opportunity to develop and refine your skills in the automatic speech recognition space
- Opportunity to collaborate on future publications
RESEARCH TO BE CONDUCTED
In an era where education dictates opportunity, reading development research must urgently support all learners to realise reading excellence. Despite this, the manner in which reading difficulties is largely assessed is very static, and most assessments are unable to distinguish between those who have a genuine reading difficulty and those who only need additional exposure to English (e.g., having English as an additional language). Moreover, given the time constraints that teachers have, an easier approach to administering these assessments is urgent. Therefore, the project seeks to investigate the effectiveness of a gamified dynamic reading screening tool administered through an online learning platform with automatic speech recognition capabilities.
One of the approaches taken in the current dynamic screening tool is to administer nonsense words (e.g., gop, beft, stendle) as a way of assessing a child’s pronunciation ability. However, most automatic speech recognition systems find it difficult to understand nonsense words (out-of-vocabulary words), with poor accuracy rates from current well-established speech systems. Therefore, a better understanding is required of how to develop speech recognition capabilities that have the flexibility of accurately and efficiently recognising these nonsense words.
You will engage in research to develop a novel algorithm and/or modelling technique to build and advance spoken language understanding systems. Within the context of the automatic speech recognition space, you will:
- Research the latest modelling techniques, including deep neural networks (DNN), episodic modelling, attention-based techniques and (hierarchical) sparse recovering modelling. The focus here will be on gaining a better understanding of the strengths and limitations of competing approaches.
- Identify modelling techniques that are likely to be most beneficial in identifying speech at the phonetic, syllabic and whole-word levels.
- Conduct mini experiments to assess the quality of speech recognition models and to assess the effectiveness of different modelling techniques.
SKILLS WISH LIST
If you’re a PhD student and meet some or all the below we want to hear from you. We strongly encourage women, indigenous and disadvantaged candidates to apply
We welcome applicants with a background in:
- Machine learning
- Statistical modelling
- Speech processing
- Machine translation
- Coding/programming, including Python, Java and/or C++
- Acoustic and language modelling
- End-to-end modelling
The overall outcome of the project involves proposing state-of-the-art and novel acoustic and language modelling methods and architectures for end-to-end learning. More specifically, the current goal is to develop a speech recognition system that is able to validly and reliably detect when it does not know a word. As part of this current project, Cinglevue are therefore seeking a speech recognition system proof of concept that is flexible in accurately and efficiently identifying a wide range of different word types. This proof-of-concept should also demonstrate capabilities that:
- Allow the recognition of out-of-vocabulary words;
- Transcribe words phonetically when the system is uncertain of a presented speech signal;
- Allow users to add new vocabulary to the speech recognition system without any technical knowledge;
- Offer sufficiently robust speech detection in noisy environments (e.g., primary school classroom setting) and
- Outperform current speech recognition approaches.
The intern will receive $3,000 per month of the internship, usually in the form of stipend payments.
It is expected that the intern will primarily undertake this research project during regular business hours, spending at least 80% of their time on-site with the industry partner. The intern will be expected to maintain contact with their academic mentor throughout the internship either through face-to-face or phone meetings as appropriate.
The intern and their academic mentor will have the opportunity to negotiate the project’s scope, milestones and timeline during the project planning stage.
CONNECT WITH APR.INTERN