# 使用 Arduino NANO 33 SENSE BLE 結合 TinyML 實現鼾聲偵測 實作目標 * 使嵌入式系統實現推論能力 * 建立神經網路模型 ![](https://i.imgur.com/om6HDVN.jpg) ## 實作步驟 1. 註冊[Kaggle](https://www.kaggle.com)與[Edge Impulse](https://www.edgeimpulse.com)帳戶 2. 從Kaggle下載分類好的鼾聲資料集 [連結-Dataset by T. H. Khan](https://www.kaggle.com/datasets/tareqkhanemu/snoring) 4. 建立新專案，接下跟著引導設定 ![](https://i.imgur.com/zBhEzVe.png =560x250) 5. 上傳資料集至Edge Impulse，自訂標籤 ![](https://i.imgur.com/yIUUEgJ.png =560x350) 6. 建立Edge Impulse 模型，Processing block -> MFCC，Learning block -> NN Classifier，Save Impulse ![](https://i.imgur.com/qEPW8f9.jpg =570x270) 7. Edge Impule MFCC設定和結果 ![](https://i.imgur.com/RYP2MJB.png =570x270) 8. Edge Impulse MFCC特徵產生 ![](https://i.imgur.com/v97cEtI.jpg =570x270) 9. 訓練模型，查看結果 ![](https://i.imgur.com/1IQE6lt.png =370x270)![](https://i.imgur.com/51TKyY6.jpg) 10. 下載訓練完的模型 ![](https://i.imgur.com/VAbF72A.png =470x370) 11. 匯入.zip檔至Arduino IDE 的 Library ![](https://i.imgur.com/0pA5lXB.png) 12. 把程式燒入至Arduino NANO 33 SENSE BLE ## 參考資料 [[Day 20] Edge Impulse + BLE Sense實現喚醒詞辨識(上)](https://ithelp.ithome.com.tw/articles/10277682) [[Day 21] Edge Impulse + BLE Sense實現喚醒詞辨識(中)](https://ithelp.ithome.com.tw/articles/10277700) [[Day 22] Edge Impulse + BLE Sense實現喚醒詞辨識(下)](https://ithelp.ithome.com.tw/articles/10278171)