first detected beat backtracking

[wwerkk]

maybe beat tracking generally should be handed over to streamer.py instead of being included in the training script?

[wwerkk]

seems like the beat tracking itself is not as important, more crucially, the silence at the beginning of the audio file should be truncated. not sure if this can be done last minute as the deadline is coming up in about 60 hours.