Vehicular communications present several challenges, especially in wireless channel estimation to ensure communications reliability in a highly dynamic environment, where many machine learning (ML) based estimator proposals can be found in the literature. However, few explore the possibility of generalizing methods for variations in vehicular channel models, which is crucial for future applications. It is common to train a model to work in specific channel conditions, such as a particular speed, fading, or modulation order. This can lead to poor performance when the system operates in different scenarios. To address this problem, this work explores the possibility of generalizing the vehicular channel estimation for different channels employing the Ensemble Learning (EL) technique, applied to a channel estimation method based on the long-term memory network (LSTM). Our results present that a generalized learning architecture can be used to estimate vehicular channels under varying conditions, resulting in an estimator that is robust to changes in Doppler-delay characteristics across different environments and channel models. Moreover, our approach achieves this generalized model without any added online complexity once the algorithm used to combine the model information is performed offline.