<![CDATA[The development of voice recognition technology, particularly Keyword Spotting (KWS) , offers significant potential for critical applications such as emergency response systems in the healthcare sector. The use of cloud-based Application Programming Interfaces (APIs) like the Google Speech API provides an accessible development pathway for these applications , but its performance in specific contexts, such as for the Indonesian language, requires thorough evaluation. This study aims to implement and evaluate the performance of a real-time KWS system based on the Google Speech API within a simple smart assistant prototype designed for healthcare emergency scenarios. The system was implemented on Google Colaboratory using a Text-based KWS approach. Quantitative testing was conducted with two male participants who uttered five Indonesian target keywords ("Hi", "Tolong", "Aduh", "Oke", "Salam") , with 25 repetitions for each keyword, resulting in a total of 250 test samples. System performance was measured using the Success Rate metric, calculated from the classification of True Positives (TP) and False Negatives (FN). The results show that the system was functionally implemented with a combined detection success rate of 62.8%. Significant performance variation was found both among keywords (highest success rate for 'Tolong' at 72.0% ) and between participants (77.6% vs. 48.0% ), highlighting speaker dependency. It is concluded that the Google API-based KWS approach is feasible for rapid prototyping, but the current accuracy and performance variability indicate the need for more robust and specific model development before it can be relied upon for critical healthcare applications.]]>
