Atomic force microscope infrared spectroscopy (AFM-IR) can perform IR spectroscopic chemical identification with sub-100 nm spatial resolution, but is relatively slow due to its low signal-to-noise ratio (SNR). In AFM-IR, tunable IR laser light is incident upon a sample, which results in a rise in temperature and thermomechanical expansion of the sample. An AFM tip in contact with the sample senses this nanometer-scale photothermal expansion. The tip motion induces cantilever vibrations, which are measured either in terms of the peak-to-peak amplitude of time-domain data or the integrated magnitude of frequency-domain data. Using a continuous Morlet wavelet transform to the cantilever dynamic response, we show that the cantilever dynamics during AFM-IR vary as a function of both time and frequency. Based on the observed cantilever response, we tailor a time-frequency-domain filter to identify the region of highest vibrational energy. This approach can increase the SNR of the AFM cantilever signal, such that the throughput is increased 32-fold compared to state-of-the art procedures. We further demonstrate significant increases in AFM-IR imaging speed and chemical identification of nanometer-scale domains in polymer films.