What are the current accuracy and limitations of the state-of-the-art automatic music transcription systems, approaches and algorithms?

I would like to answer this question from a research perspective. A lot of work has been done on this topic and this problem has been solved in several stages.
1. Field Determination: It has achieved quite good precision by using different algorithms such as YIN by its name.
2. Speed ​​Detection: This area has its own challenges, as you look at the type of music you intend to enter.

Context-based optimization to correct any diagnosis prediction
In general, they have all been tested as offline models and yielded good results for audio tracks that have the same device.




The name of the field you want to search in is called Music Info Search.

Recognizing the accent is still early, but you will get good results on a product like Melody.

If you want to go practically, I suggest you try as much / a teaspoon or audio focus frame as the audio.

You will also want to look at auditory scene analysis and spectrum morphology, two very useful tools for defining acoustic properties.

I have done a number of projects that have focused on trans-rhythm classification and transient detection for suicidal transcription, and I have found that the sound function of the profile is F. The change of horror is very useful.

I have never looked too much at the identity of the accent, but I do think that the solar wave will start to find a way to periodically identify.