Multi-Resolution Analysis (MRA) is a family of techniques of time-frequency analysis that make different time-frequency resolution trade-offs for different frequencies. Conventional Short-Time Fourier Transform (STFT, including its most common implementation Fast Fourier Transform) picks a fixed window size, resulting in fixed resolution. MRA gets around this by effectively picking large window sizes for lower frequencies and small window sizes for higher frequencies.
When applied to voice signals, MRA is thus able to visually represent both the persistent energy near the fundamental frequency and the transient, pulsing events--individual glottal impulses and the vocal tract resonance responses.
Wavelet Transform is a class of widely-used techniques that embeds the idea of MRA in its algorithm: it convolves a wavelet function that is scaled according to the particular frequency to extract information for that frequency. This website uses the waipy library under the hood to run Continuous Wavelet Transform and generate the above scaleograms.
Xiao Shi gave a short talk on MRA on the singing voice at the Voice Foundation Symposium in June 2022, the slides are available here.