Presentation Information
[8p-N304-5]Spoken Digit Classification Using a Transmon Quantum Reservoir
〇(D)Ken Arita1, Edmund Soji Otabe1, Yuki Usami1, Ahmet Karakari1, Muzhen Xu1, Hirofumi Tanaka1, Tetsuya Matsuno2 (1.Kyushu Inst. Tech., 2.NIT-Aritake)
Keywords:
reservoir computing,transmon qubit,spoken digit classification
Spoken digit classification was performed using quantum reservoir computing based on transmon qubits. Speech signals were divided into frequency bands and encoded as inputs to a quantum system, where reservoir dynamics were generated through time evolution. Expectation values extracted from the evolved quantum states were used as features for classification. The results showed that even a two-qubit reservoir with only 32 features achieved an accuracy of 85%. Furthermore, under a fixed feature budget, the configuration with four qubits and eight temporal readout points achieved the highest accuracy of 90%. These findings indicate that quantum reservoir computing can effectively generate useful features for speech recognition and that an appropriate balance between the number of qubits and temporal nodes is crucial for achieving high classification performance.