To optimize the noise that heard like ‘kalakala’ produced by the plug hybrid electric vehicle when accelerating with a small accelerated pedal opening while in the charging state of series modal. The LMS test device was used to acquire the noise of the driver's outer ear. Through filtering and playback analysis, it was confirmed that the noise is mainly contains the frequency bands of 250-400Hz and450-700Hz. The frequency bands of the noise were used as carriers for Hilbert transform, and their envelopes were obtained for Fourier transform analysis. It was found that the modulation order of the noise is 0.5 times of the engine ignition order, and the modulation frequency is 20-30Hz, which let the customer hears like roughness. Regarding the spectral characteristics of this noise, firstly, at the excitation source, selected a reasonable moment of inertia and frequency of the Crank torsional damper, to decrease the torsional excitation of the engine. Secondly, investigated the structural propagation path of the noise systemically, and effectively reduces the propagation path of the noise by increasing the dynamic stiffness of the powertrain mount’s passive bracket and tuning a reasonable front driveshaft damper. Finally, based on the universal characteristics and the pressure rise rate map of the engine, as well as the targets of the attribute of the vehicle economy and power, different calibration strategies were attempted, and a calibration strategy that balances the three attributes was selected. By the above measures, the energy of two resonance bands of acceleration roughness sound were reduced by 9dB (A), and 3dB (A) respectively. After optimization, the noise inside the car is almost inaudible.