<div>Speech enhancement (SE) systems aim to improve the quality and intelligibility of degraded speech signals obtained from far-field microphones. Subjective evaluation of the intelligibility performance of these SE systems is uncommon. Instead, objective intelligibility measures (OIMs) are generally used to predict subjective performance increases. Many recent deep learning based SE systems, are expected to improve the intelligibility of degraded speech as measured by OIMs. </div><div><br></div><div>However, validation of the OIMs for this purpose is lacking. Therefore, in this study, we evaluate the predictive performance of five popular OIMs. We compare the metrics' predictions with subjective results. For this purpose, we recruited 50 human listeners, and subjectively tested both single channel and multi-channel Deep Complex Convolutional Recurrent Network (DCCRN) based speech systems. </div><div><br></div><div>We find that none of the OIMs gave reliable predictions, and that all OIMs overestimated the intelligibility of `enhanced' speech signals. </div>

Hearing loss may be mild, moderate, severe or profound and can affect one or both ears. Without a systematic approach of detecting HL, only those with more severe HL are detected, often by the community (guardians, teachers, health workers and peers). This means that persons with mild to moderate HL often go undetected, even if such HL still leads to difficulty in hearing conversational speech (World Health Organization 2020). As listening is a main form of learning, children with HL often have lower school performance than children without HL (Flexer, Millin & Brown 1990;Lieu et al. 2010). In many LMICs, children with HL and deafness are vulnerable to dropping out of school, not achieving expected learning goals or never going to school, with girls being more at risk of dropping out or never attending (Njelesani et al. 2018;UNICEF n.d.; World Background: Hearing is essential for learning in school, and untreated hearing loss may hinder quality education and equal opportunities. Detection of children with hearing loss is the first step in improving the learning situation, but effective interventions must also be provided. Hearing aids can provide great benefit for children with hearing impairment, but this may not be a realistic alternative in many low-and middle-income countries because of the shortage of hearing aids and hearing care service providers.Objective: In this study, alternative solutions were tested to investigate the potential to improve the learning situation for children with hearing impairment.Method: Two technical solutions (a personal amplifier with and without remote microphone) were tested, in addition to an approach where the children with hearing impairment were moved closer to the teacher. A Swahili speech-in-noise test was developed and used to assess the effect of the interventions. Results:The personal sound amplifier with wireless transmission of sound from the teacher to the child gave the best results in the speech-in-noise test. The amplifier with directive microphone had limited effect and was outperformed by the intervention where the child was moved closer to the teacher. Conclusion:This study, although small in sample size, showed that personal amplification with directive microphones did little to assist children with hearing impairment. It also indicated that simple actions can be used to improve the learning situation for children with hearing impairment but that the context (e.g. room acoustical parameters) must be taken into account when implementing interventions. Contribution:The study gives insight into how to improve the learning situation for school children with hearing impairment and raises concerns about some of the known technical solutions currently being used.

<div>Speech enhancement (SE) systems aim to improve the quality and intelligibility of degraded speech signals obtained from far-field microphones. Subjective evaluation of the intelligibility performance of these SE systems is uncommon. Instead, objective intelligibility measures (OIMs) are generally used to predict subjective performance increases. Many recent deep learning based SE systems, are expected to improve the intelligibility of degraded speech as measured by OIMs. </div><div><br></div><div>However, validation of the OIMs for this purpose is lacking. Therefore, in this study, we evaluate the predictive performance of five popular OIMs. We compare the metrics' predictions with subjective results. For this purpose, we recruited 50 human listeners, and subjectively tested both single channel and multi-channel Deep Complex Convolutional Recurrent Network (DCCRN) based speech systems. </div><div><br></div><div>We find that none of the OIMs gave reliable predictions, and that all OIMs overestimated the intelligibility of `enhanced' speech signals. </div>