artem4852 / piano-transcription-project

What can this application do?

This application can be used to make the process of transcription of an audio file with piano music much quicker and easier. It uses the power of machine learning to predict information about each note, such as its pitch, length, and information about the rests after it. It also has an interface to easily correct the pitch predictions made by the application. It allows users to then export the transcription as an mxl or midi file, which can be opened in music notation software such as MuseScore or Finale.

---

Important to note

The application predicts the notes "one at a time", so no chords and only one (treble) clef kind of music could be predicted. Since the algorithms were trained using audio data from MuseScore 4, the application would work best with audio files from that app. It might not work completely well with audio files from other apps, because of slight variations in pitch and volume.

Also, the application might not spot the first note if it comes from the very first second in audio file. This is because the application needs to have some audio data before the note to be able to recognise and separate it. In the testing files, the whole rest was added to first measure before the notes to prevent this issue.

---

How to install the application

First, download the file and unzip it.

Then, on Windows, open the powershell and navigate to the folder where you unzipped the file. Then, run the following command:

pip3 install -r requirements.txt

On Mac and Linux, open the terminal and navigate to the folder where you unzipped the file. Then, run the following command:

pip3 install -r requirements.txt && chmod +x ./start.sh

The application also requires you to have ffmpeg installed. Please use the tutorials to install it if you haven't already. For Windows I suggest to use this tutorial and for Mac this one.

---

Usage

Starting the application

On Windows, open the powershell and navigate to the folder where you unzipped the file. Then, run the following command:

python3 main.py

On Mac, open the terminal and navigate to the folder where you unzipped the file. Then, run the following command:

sudo ./start.sh

On Linux, open the terminal and navigate to the folder where you unzipped the file. Then, run the following command:

sudo -E python3 main.py

Choosing the language

When you first run the file, you would be prompted to choose the language. You can choose between English, Ukrainian and russian. To choose the language, type in the number of the language you want to choose and press enter.

Transcribing a file

After the models are loaded, you would need to input the path to the audio file that should be transcribed. Then, the application would split the file into notes, extract features, and predict the data about each note and rest.

Correcting the pitch

After the transcription is done, you could choose to correct the pitch of the notes. After entering the UI, the small instructions would appear below the list of note predictions. To navigate between the notes, left/right arrow keys are used. To hear how note sounded in the file, press "O", to hear how the note sounds currectly press "C". When navigating between the notes you would also hear how the note sounds currently. This makes it very easy to see it the pitches match or not. If they don't use the up/down arrow keys to change the pitch. After you are done, press "esc" to exit the UI.

Exporting the transcription

In the end, you can export the transcription as an mxl or midi file. You would first be prompted to enter the format you want to export the file as. Then, you would be prompted to choose where to export the file (the options are: the "results" folder of the application, the same place as the audio file is in, and custom path). If you choose the "custom path" option, you would then be prompted to enter it. After that, the file would be exported. The file would be named the same as the audio file. It could be opened in music notation software such as MuseScore or Finale.

---

Feedback

Your feedback is very important to me as a developer. Please fill in this form to give me feedback about the application. It would only take a minute of your time.

---

Errors

If you encounter any errors, please create an issue. Please include the steps to reproduce the error, the error message, your operating system, and the version of python you are using.