Contemporary automated speech analysis systems face several fundamental limitations, including the fragmented analysis of linguistic and emotional characteristics, a focus on a limited set of basic emotions, and insufficient adaptation to diverse language systems. This study proposes a comprehensive approach to speech feature analysis, consisting of two complementary components designed to overcome these limitations. The proposed framework comprises two complementary methods. The first method is designed to assess the level of speech activity and is based on a comprehensive analysis of 36 linguistic parameters. These parameters include quantitative features (mean utterance length, number of sentences), syntactic features (construction complexity), lexical diversity (vocabulary richness), as well as the frequency of various parts of speech. The MannWhitney U test is employed to classify the level of speech activity. The second method is a stateof-the-art multi-task neural network architecture based on the pre-trained RuBERT-large language model. This architecture is capable of simultaneously evaluating three fundamental parameters of emotional state: valence, arousal, and dominance. The combination of these parameters enables the identification of 26 complex emotional states. Experimental validation on heterogeneous Russianand English-language corpora, including professional speech and spontaneous dialogues, demonstrated the superiority of the proposed methods over existing analogues. The speech activity assessment method achieved an accuracy of 92% for English and 89% for Russian. The multi-task model attained an accuracy of 85% in determining valence, 80% for arousal, and 76% for dominance. For 10 main emotional categories, the classification accuracy reached 57%. The results of this study can be applied to the development of intelligent dialogue systems and chatbots capable of adapting their communication style and emotional responses based on the user's speech activity and emotional state, thereby significantly enhancing the quality and naturalness of interaction.