bark Output doesn't match demos

Hi I ran this locally

from bark import SAMPLE_RATE, generate_audio
from IPython.display import Audio

text_prompt = """
     ♪ In the jungle, the mighty jungle, the lion barks tonight ♪
"""
audio_array = generate_audio(text_prompt)
Audio(audio_array, rate=SAMPLE_RATE)

Play the file: https://user-images.githubusercontent.com/7272343/233417746-dbf0ab65-49c7-477c-9373-1b4f87bdfb5e.mp4

Very odd sound any ideas?

Apr 20 '23 15:04 abacaj

hm, yeah that sounds like a dud. try generating a few more times and see what it sounds like? it's an autoregressive model like GPT, so sometimes it goes off the rails. especially music can sometimes cause that cause rhythm increases the chances of a weird loop. We're working on improving things but for now you can just run it a few times till you get something good probably

Apr 20 '23 15:04 gkucsko

This one came out pretty funny, almost like it didn't wanna sing

Sample: https://user-images.githubusercontent.com/7272343/233423797-3f6f45eb-65c5-4e0b-b656-0714c92a3821.mp4

Apr 20 '23 16:04 abacaj

😂

Apr 20 '23 16:04 gkucsko

In our experiments with this set of models, for more interesting tokens such as ♪, you might need several generations to get something very good, but the good generations are very good.

Apr 20 '23 20:04 mcamac