Odyssey 2024 Emotion Recognition Challenge aims to compare different emotion recognition systems in two tasks: classifying speech across eight emotional classes and predicting emotional attributes for arousal, valence and dominance. This paper describes TalTech’s systems prepared for the challenge that fuse the predictions of text and speech based emotion recognition models. The audio-based model adapts the Wav2Vec2-BERT model for emotion recognition, while the text-based model uses finetuned LLaMA2-7B as the backbone. The two models are combined for the classification task by training a multi-class logistic regression model, using the posteriors of the underlying models as input features. The model obtained a macro-F1 score of 0.354 on evaluation data and was ranked 2nd among all teams. The fusion model for the attribute prediction task achieved an average score of 0.5144 and was thereby ranked 6th among the teams.