This paper proposes two simulation techniques of the Lombard effect. These simulations are based on the Linear Prediction Coding (LPC) and the Linear Multiple Regression (LMR) methods. The LPC model determines a synthesis filter. The LPC coefficients filter are obtained by the processing of the spectral noise signal This transformation simulates both enhancement spectral tilt and a relative amplification of speech spectral frequency bands where maximum of noise energy exists. This approximates the Lombard effect. These experiments are used to test the limits of SAMREC1 (ENST's DTW recognizer system) in presence of Lombard effect. The second technique is based on the learning of the spectral transformation from the database reference without noise to the same database but recorded with the Lombard effect simulated in the laboratory. This treatment is optimized by dynamic programming.