[23p-A302-1~13] FS.1 Focused Session "AI Electronics"

3:45 PM - 4:00 PM JST(6:45 AM - 7:00 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation AwardEnglish Presentation

[23p-A302-10] Speech Recognition with Time-Delayed In-Materio Reservoir Computation

〇(M2)Ahmet KARACALI¹, Oradee Srikimkaew¹, Yuki Usami^1,2, Hirofumi Tanaka^1,2 (1.LSSE, Kyushu Inst. Technol. (Kyutech), 2.Neuromorphic Center, Kyutech)

Keywords:

Reservoir Computing,Speech Recognition,MFCC

The physical RC device performs more efficiently in reservoir computation than conventional neuron network systems in terms of power consumption. In this study, we used this physical reservoir device for speech classification. Firstly, the audio files in the free-speech cascaded dataset were converted into electrical signals. Then, time series data were collected with the time-delayed reservoir device by applying the MFCC (Mel-Frequency Cepstral Coefficients) feature selection algorithm. This time series data was trained and speech classification was performed.