<![CDATA[Human–computer interaction is increasingly evolving toward more natural and efficient voice-based communication. However, most voice assistant systems still separate speech recognition from users’ emotional analysis, resulting in less adaptive interactions. This study aims to design and implement an integrated speech recognition system that combines Speech-to-Text (STT), Support Vector Machine (SVM)-based sentiment analysis, and Text-to-Speech (TTS) within a unified voice-to-voice pipeline. The system was developed using the prototype method to ensure stable and iterative integration among the modules. The STT module utilizes the Google Web Speech API for speech transcription, while sentiment classification employs the SVM algorithm supported by preprocessing stages, including text normalization and spelling correction. The results show that the system operates in real time with a stable response time ranging from 0.6 to 0.9 seconds. Evaluation of the STT module using the Word Error Rate (WER) metric demonstrated optimal performance, achieving a WER of 0 on the test data. In sentiment analysis testing, the prototype method significantly improved system accuracy from 56.00% in the initial prototype to 85.33% in the final prototype through the addition of training data and model refinement. The integration of the three components using the Flask framework resulted in a virtual assistant capable not only of converting speech into text but also of responding adaptively to users’ sentiments through voice. In conclusion, integrating STT, SVM, and TTS into a unified pipeline effectively improves the quality of voice interaction, enabling more communicative and adaptive human–computer interaction.]]>
