Audio data format

[Freecheck100]

Thanks for your work.

I followed your guidance on README to prepare for the MUSIC-AVQA dataset. But when dealing with the audio part, what I get is the format of wav, not the mp3 format as required in the file music_avqa_valid_train_samples.json.

So I wonder What the problem is? If I just need to compress the wav data to the mp3 format, what is the bitrate?

Looking forward to your reply.

[echo0409]

Thanks for your interest!

Both formats (wav and mp3) can work well. You can directly use the wav format.

[Freecheck100]

Thanks for your reply! I also wonder the checkpoint you choose for the benchmark. After training, there are so many checkpoints saved as your script does, but which is used? Is the last checkpoint "checkpoint-11970" always used for every benchmark?

[echo0409]

Thanks for your reply! I also wonder the checkpoint you choose for the benchmark. After training, there are so many checkpoints saved as your script does, but which is used? Is the last checkpoint "checkpoint-11970" always used for every benchmark?

For the AVL case, we use the last checkpoint for evaluation :)

[Freecheck100]

Got it! I also wonder could you please give the test_samples.json of MUSIC-AVQA? I will really appreciate it.