Although pulsed coherent laser radar vibrometry has been introduced as an improvement over its Continuous Wave (CW) counterpart, it remains very sensitive to decorrelation noises such as speckle, and other disturbances of the measurement. Taking advantage of more polyvalent poly-pulse waveforms, we address the issue with advanced signal processing. We have conducted what we believe is the first extensive comparison of processing techniques considering CW, pulse-pair and poly-pulse emissions. In this framework, we introduce a computationally efficient maximum likelihood estimator and test signal tracking on pseudo time-frequency representations, which respectively help deal with speckle noise and fading of the signal, in harsh noise conditions. Our comparison on simulated signals is validated on a 1.55µm all-fiber vibrometer experiment, with an apparatus simulating vibration and strong speckle noise. Results show the advantage of estimators taking into account actual noise statistics, and call for a wider use of time-frequency representations to track the vibration modulated signal.