Hardware–software co-design of an audio feature extraction pipeline for machine learning applicationsVreča, Jure (Avtor) Pilipović, Ratko (Avtor) Biasizzo, Anton (Avtor) FPGAMFCCkeyword spottingchiselKeyword spotting is an important part of modern speech recognition pipelines. Typical contemporary keyword-spotting systems are based on Mel-Frequency Cepstral Coefficient (MFCC) audio features, which are relatively complex to compute. Considering the always-on nature of many keyword-spotting systems, it is prudent to optimize this part of the detection pipeline. We explore the simplifications of the MFCC audio features and derive a simplified version that can be more easily used in embedded applications. Additionally, we implement a hardware generator that generates an appropriate hardware pipeline for the simplified audio feature extraction. Using Chisel4ml framework, we integrate hardware generators into Python-based Keras framework, which facilitates the training process of the machine learning models using our simplified audio features.MDPI20242024-03-25 10:55:57Neznano18556UDK: 004ISSN pri članku: 2079-9292DOI: 10.3390/electronics13050875COBISS_ID: 186803203Švicasl© 2024 by the authors.