Several clarinetists’ judgment of the automatic evaluation system for clarinet reed’s playability

The clarinet produces sound by means of a vibrating reed, typically made from cane and sold in sets of 10. Each set contains them with varying degrees of playability. It has been reported that on average approximately only 3 among the 10 are possible to use to give performances in concerts (Suzuki et al., 2024). The playability of a reed is critical that affects not only the quality of sound produced, but also performance technique with overall motivation for clarinetists. The playability of a reed can only be judged by just trying it. Even clarinet majors at music colleges may not make the correct judgment. Some visual characteristics of a reed offer insights into its playability without playing it based on clarinetists’ experiences. The authors developed a computational system that automatically evaluates the playability of clarinet reeds from images using a convolutional neural network. Reeds were photographed using both reflected and transmitted light, then the system was trained based on the first author's evaluations of playability. The system achieved an accuracy of 0.90 and an F-measure of 0.88. We describe how the system was used by different players and present their impressions and the corresponding results.