Part2_Music_Generation: mdl.lab1.extract_song_snippet(generated_text) outputs "Found 0 songs in text"

[rvh72]

In Part2_Music_Generation.ipynb, I get an unexpected output from the API to generate music from predicted notes. This is what I see:

generated_songs = mdl.lab1.extract_song_snippet(generated_text)
print(generated_songs)

Found 0 songs in text
[]

The solutions notebook has the same behavior. The value of is valid, see uploaded txt file:

print(generated_text)

generated_txt.txt

Any idea what is going wrong?

[tspeterkim]

Just try regenerating the generated_text a couple times.

There is some element of chance as the generated text might not be a valid song in text or syntax.

[aamini]

The generated text file you shared is not a valid song so the error message you are seeing is correct. There are several reasons for this:

1. You need to train the network longer to fit the songs better
2. You need to regenerate a new sample song since there is some randomness in the sampling (as @peterkim95 indicated above)

[Raunak-Singh-Inventor]

I found that the songs are only generated if you are on Google Colab. When working on local jupyterlab, it isn't able to generate the song even if it is valid.

[RudrakshSugandhi]

@aamini said to try to increase the num_iteration from 2000 to 4000 or more

num_training_iterations = 4000

Then it will generate the songs.

[Praveen880890]

try a different start string like "AVCV" or something longer instead of just "X" when generating the song