Earlier work has shown tool failure monitoring using frequency-based pattern recognition analysis of acoustic emission signals to be feasible while machining under fixed cutting conditions. However, cutting conditions change quite frequently in industrial production, and since AE signals are affected by varying conditions, a model is developed based on Taylor's expansion using experimental data obtained for various process variables and their output AE spectra, and is used to filter the influence of varying conditions on signal classification. The experimental study involved a fractional factorial experimental design which delineated effects of variables on AE generation during machining. Normalized AE spectra within the 100 to 1000 kHz range were used as system output along with the AE power. While the normalized spectra were found unaffected by changes in the depth of cut, the total AE power was also little affected by the mixed flank and crater wear, and feed rate changes. Using the model and filter designed, performance of the tool wear, chip noise and tool breakage classifier improved to 83, 99, and 97 percent classification, respectively.