Volume - 7 | Issue - 4 | december 2025
Published
04 November, 2025
Human-computer interaction (HCI) applications increasingly rely on reading speech, understanding emotional context, and generating natural language. The heterogeneous set of approaches the existing solutions use for sentiment analysis, speech synthesis, and speech recognition results in an unbalanced user experience. Designing an integrated system that can perform speech-to-text (STT) and text-to-speech (TTS) processing, sentiment analysis of input text, and neighboring aware speech generation is the problem this paper attempts to solve. To identify complexity like sarcasm and negation, Bi – LSTM is selected because it can learn context from context words (previous and next words in a sentence). In spite of data sparsity conditions, GloVe embeddings improve model generalisation by offering deep semantic understanding from large corpora. Following experimental verification, our Bi-LSTM with GloVe embeddings achieves 90% sentiment classification accuracy that is 7-10% higher relative to standard baselines like SVM (82%) and Naïve Bayes (75%). With true positive values above 88%, the model achieves well-balanced performance on the positive, neutral, and negative classes. Due to its low latency and about 87% accuracy during live testing, the system is an excellent option for interactive systems. All these features are amalgamated in our Phonetic Flow System, which enhances them to develop an extensible system that supports quicker, more natural, and emotionally intelligent human-machine interaction.
KeywordsHCI STT TTS Bi-LSTM.
