A Matlab App that extracts the partition of a music by studying its temporal-frequential representation (spectrogram). This is certainly the processing done by Shazam in order to recognize a music. This app enables to:
- Study Temporal/Frequential and Spectrogram representation of a signal
- Extract the partition of the music
- Regenerate the music using this partition
- Temporal + Frequential Analysis enable to determine the best compromise of accuracy.
↗ Frequency Accuracy ⇒ ↘ Temporal Accuracy. Need both to analysis a music. - Compromise chosen → 21Hz of frequency resolution and 0.046s of temporal resolution
- 20Hz is enough in order to regroup close notes into the same frequency
- Using an appropriate threshold enables to exctract main notes while removing a maximum of noise
- The chosen algorithm for the extraction of the partition is a flood-fill algorithm (homemade)
- This enables to avoid doublons of occurences of notes
- Based on the partition and the power of each note, it is possible to regerate the music.
- You can hear the difference in the example in the App:
between
lac_des_cygnes_original.wavandlac_des_cygnes_encoded.wav