This paper presents a methodology for analyzing polyphonic musical passages, particularly those with notes that exhibit a harmonically fixed spectral profile, such as piano notes. The approach leverages the unique note structure to model the audio content using a linear basis transform and employs non-negative matrix factorization (NMF) to estimate the spectral profile and temporal information of each note. This method results in a simple, compact system that learns from data rather than relying on prior knowledge. The authors propose a data-driven, redundancy reduction approach similar to scene analysis, which has shown promising results in polyphonic music transcription. NMF is used to decompose the magnitude spectrogram of the musical passage into two non-negative matrices, $\mathbf{W}$ and $\mathbf{H}$, where $\mathbf{W}$ represents the spectral profiles and $\mathbf{H}$ represents the temporal information. The rank $R$ of the approximation controls the level of summarization, with smaller values of $R$ providing more detailed descriptions. The paper demonstrates the effectiveness of NMF on real piano recordings, including isolated notes and coinciding notes. For isolated notes, NMF accurately identifies the individual notes and their frequencies. For coinciding notes, the method sometimes consolidates multiple notes into a single component due to the system's learning process, but this can be mitigated by providing enough data to highlight the individuality of each note. The authors conclude that their NMF-based approach for polyphonic music transcription is efficient and effective, but it requires music passages from instruments with notes that have a static harmonic profile. Future work will explore alternative decomposition methods to address this limitation.This paper presents a methodology for analyzing polyphonic musical passages, particularly those with notes that exhibit a harmonically fixed spectral profile, such as piano notes. The approach leverages the unique note structure to model the audio content using a linear basis transform and employs non-negative matrix factorization (NMF) to estimate the spectral profile and temporal information of each note. This method results in a simple, compact system that learns from data rather than relying on prior knowledge. The authors propose a data-driven, redundancy reduction approach similar to scene analysis, which has shown promising results in polyphonic music transcription. NMF is used to decompose the magnitude spectrogram of the musical passage into two non-negative matrices, $\mathbf{W}$ and $\mathbf{H}$, where $\mathbf{W}$ represents the spectral profiles and $\mathbf{H}$ represents the temporal information. The rank $R$ of the approximation controls the level of summarization, with smaller values of $R$ providing more detailed descriptions. The paper demonstrates the effectiveness of NMF on real piano recordings, including isolated notes and coinciding notes. For isolated notes, NMF accurately identifies the individual notes and their frequencies. For coinciding notes, the method sometimes consolidates multiple notes into a single component due to the system's learning process, but this can be mitigated by providing enough data to highlight the individuality of each note. The authors conclude that their NMF-based approach for polyphonic music transcription is efficient and effective, but it requires music passages from instruments with notes that have a static harmonic profile. Future work will explore alternative decomposition methods to address this limitation.