Using audio devices to collect speech from far is a significant and challenging task. The speech energy severely decays if the speaker is far away from the audio acquisition device, resulting in poor speech quality. Usually, a long-range speech acquisition task can be seen as a speech enhancement problem in a specific application scenario. We proposed a speech enhancement method called Paraboloid with Microphone Array (PMA) to enhance long-range speech. Firstly, a parabolic reflector is employed to enhance the target signal acoustically. Meanwhile, the Linearly Constrained Minimum Variance(LCMV) algorithm based on a microphone array is used to jam the target speech and estimate the noise. Then a mapping relationship between the above two output noises is established using an improved Long Short-Term Memory(LSTM) neural network. At last, with the mapping model, the final enhanced speech is obtained by reducing the LCMV estimated noise from the acoustically enhanced speech remained noise. Computer simulations show that the proposed method can effectively enhance speech from the speaker within 50 meters. Besides, the outdoor experiment with a realistic PMA and the subjective listening test also confirmed the effectiveness of the PMA in a real-world scenario.