e-Learning is becoming an increasingly popular educational paradigm because of the rapid growth of the Internet. Recent studies have argued that affective modelling (ie, considering a learner's emotional or motivational state) should also be considered while designing learning activities. Many studies indicated that various learning emotions markedly impact learning outcomes. In the language education field, many studies have investigated anxiety associated with learning a second language, noting that anxiety has an adverse effect on the performance of those speaking English as a second language. Therefore, how to reduce anxiety associated with learning a second language to increase learning performance is an important research issue in the language education field. Accordingly, this study employed a sensor, signal processing, wireless communication, system-on-chip and machine-learning techniques in developing an embedded human emotion recognition system based on human pulse signals for detecting three human emotions-nervousness, peace and joy-to help teachers reduce language-learning anxiety of individual learners in a web-based one-to-one synchronous learning environment. The accuracy rate of the proposed emotion recognition model evaluated by cross-validation is as high as 79.7136% when filtering out human pulse signals that have bias. Moreover, this study applied the embedded emotion recognition system to assist instructor's teaching in a synchronous English conversation environment by immediately reporting variations in individual learner emotions to the teacher during learning. In this instructional experiment, the teacher can give appropriate learning assistance or guidance based on the emotion states of individual learners. Experimental results indicate that the proposed embedded human emotion recognition system is helpful in reducing language-based anxiety, thus promoting instruction effectiveness in English conversation classes.