Automatic Transcription of Music
| Main Authors: | -, Klemens; Program Studi Teknik Informatika, -, Liliana; Program Studi Teknik Informatika, Budhi, Gregorius Satia; Program Studi Teknik Informatika |
|---|---|
| Other Authors: | Program Studi Teknik Informatika Universitas Kristen Petra |
| Format: | Article info application/pdf eJournal |
| Bahasa: | eng |
| Terbitan: |
Jurnal Infra
, 2014
|
| Subjects: | |
| Online Access: |
http://studentjournal.petra.ac.id/index.php/teknik-informatika/article/view/1440 http://studentjournal.petra.ac.id/index.php/teknik-informatika/article/view/1440/1298 |
Daftar Isi:
- Transcription of music is an active research problem in computer science and has many applications in digital music. The use of time-domain filter to improve sound quality and adding sound effects is commonplace, but the use of filter for pitch detection is rarely explored. This work thesis aims to create a pitch detection system based on digital filter and test it’s performance in guitar and piano signals. IIR Comb Filter and Peak2 onset detection is chosen as the main algorithm for this application. IIR Comb Filter have notches, zero values corresponding to a certain frequency and all it’s harmonic frequencies. If make IIR Comb Filter correspond to a pitch frequency, all its harmonic frequency will have zero values. Filtering this kind of filter with a pitched signal will result in elemination of a pitch frequency and all it’s harmonics, and the filter corresponding to the input frequency will have a minimum output compared to the other filter’s result. Peak2 onset detection works by taking the result of two peak detectors. The first peak detector detects sudden changes in amplitude in time-domain, the second peak detector takes the first peak detector’s values as an input, producing a smother result. Test showed that the result varied between instruments. Onset detection for guitar and piano performs rather poorly, about 55% of all the correct onsets were detected. For pitch estimation, the accuracy for the pitch detected is 81.25% for all monophonic signals and 64.29% for all polyphonic signals.