search

slSeanWU / MusDr

Evaluation metrics for machine-composed symbolic music. Paper: "The Jazz Transformer on the Front Line: Exploring the Shortcomings of AI-Composed Music through Quantitative Measures", ISMIR 2020
MIT License
60 stars 7 forks source link

Audio formats——.midi or .mp3 #3

Open Jessiswong opened 3 years ago

Jessiswong commented 3 years ago

Hello, thank you very much for your work, it is very meaningful, but I have some questions to seek your answers. In the inference. py in Jazz Transformer,the final generation is the audio in .midi format, but in the final presentation and objective evaluation (MUSDR) you gave all the audio in .mp3 format. Is it necessary to do this? If so, how do you implement the midi to mp3 format conversion? Looking forward to your reply.

slSeanWU commented 3 years ago

Hi,

Thanks for the message. Among the metrics, only Structureness Indicators (SI) depend on audio inputs, the others work on REMI encodings (check the symbolic/ directory).

As for midi to mp3 conversion, I use command-line synthesizer tool FluidSynth (https://www.fluidsynth.org/) and python package midi2audio (https://pypi.org/project/midi2audio/). You can also check this site: https://sites.google.com/site/soundfonts4u/ for better-sounding soundfonts.

Hope this helps!

huangqing6 commented 2 years ago

Hello，I would like to know if there is a code to convert midi to mp3. If so, can you upload it?Thank you very much!