20.500.12556/RUL-187038 Prepoznavanje govorjenih ukazov na vgrajenih sistemih Keyword Spotting on Embedded Systems Diplomsko delo obravnava prepoznavanje govorjenih ukazov na vgrajenem sistemu s pomočjo strojnega učenja. V okviru dela smo zasnovali in implementirali celoten sistem za prepoznavanje šestih govorjenih ukazov (“left”, “right”, “up”, “down”, “stop”, “go”) na razvojni plošči STM32H750B-DK. Model smo naučili na podatkovni zbirki Google Speech Commands z uporabo konvolucijske nevronske mreže tipa DS-CNN, vhodne značilke pa smo izračunali s postopkom MFCC (Mel-Frequency Cepstral Coefficients). Za zagotovitev primerne velikosti in hitrosti izvajanja na vgrajenem sistemu smo model kvantizirali v format INT8 in ga s pomočjo orodja X-CUBE-AI pretvorili v izvorno kodo C. Na vgrajenem sistemu smo implementirali celoten cevovod obdelave zvočnega signala – od zajema preko avdio vmesnika, filtriranja in prevzorčenja, izračuna MFCC značilk v jeziku C, do zagona modela in prikaza rezultata na zaslonu LCD. Naučeni model je na testni množici dosegel točnost 93,07 %, po kvantizaciji pa 92,90 %, kar potrjuje, da izguba točnosti zaradi kvantizacije ni bistvena. Rezultati kažejo, da je predlagana rešitev primerna za prepoznavanje govorjenih ukazov v realnem času na vgrajenih sistemih z omejenimi računskimi viri. This thesis addresses keyword spotting on an embedded system using machine learning methods. We designed and implemented a complete system for recognizing six spoken commands (“left”, “right”, “up”, “down”, “stop”, “go”) on the STM32H750B-DK development board. The model was trained on the Google Speech Commands dataset using a depthwise separable convolutional neural network (DS-CNN), with input features computed using the MFCC (Mel-Frequency Cepstral Coefficients) method. To achieve a suitable model size and execution speed on the embedded target, the model was quantized to INT8 and converted to C source code using the X-CUBE-AI tool. On the embedded system, we implemented the complete audio processing pipeline – from signal acquisition via the audio interface, filtering and resampling, MFCC feature computation in C, to model inference and result display on the LCD screen. The trained model achieved 93.07 % accuracy on the test set, and 92.90 % after quantization, confirming that the accuracy loss due to quantization is negligible. The results show that the proposed solution is suitable for real-time keyword spotting on embedded systems with limited computational resources. vgrajeni sistemi prepoznavanje govorjenih ukazov strojno učenje MFCC konvolucijske nevronske mreže kvantizacija embedded systems keyword spotting machine learning MFCC convolutional neural networks quantization true false false Slovenski jezik Angleški jezik Diplomsko delo/naloga 2026-09-08 13:40:06 2026-09-08 13:40:13 2026-09-17 12:06:52 0000-00-00 00:00:00 2026 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 38908 291441411 Ivanovic_Sasa_-_Prepoznavanje_govorjenih_ukazov_na_vgrajenih_sistemih.pdf Ivanovic_Sasa_-_Prepoznavanje_govorjenih_ukazov_na_vgrajenih_sistemih.pdf 1 0F05AD012FB934E3D5EDDB554500C4D8 296c03b82b4724e0a3de53aff45447ec80975d14a77a15db0c0537cd49e70eca 08250647-ab7a-11f1-8bc5-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=246471 Fakulteta za računalništvo in informatiko 0 0 0