![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
lyric generator
Apr. 23rd, 2020 11:34 amI finally found a Fiverr dev willing to take on my little lyric generator project. It's been interesting playing product owner on a teeny tiny scale. His first pass only took him a couple of hours and was pretty cool, but it had a couple defects and one of the completion criteria was completely misinterpreted, but I figure in all, that's pretty good for the couple paragraphs and sharpie wireframes I gave him.
Here's an example of some of the things that come out based on a corpus of lyrics from the band hum:
my eye to see far to spare and she miss the train.
differ from her dream with absent eye.
upright let me to come back now.
do care what they choos.
dream with seven nine and think about the end.
I guess the biggest issue I have is that he used the NLTK routine for "stemming" the words, hence some of them missing their suffixes. If this were intended to be generating grammar from scratch, that would matter, but I'm just going off of probability, so it doesn't. He said he'll pull it out in the next rev, and I think the usefulness of output will go up a lot.
I only briefly looked over the code. It's not how I would have done it / want it to be in some ways, but I guess in terms of utility, it still gets me farther than I was. For example, he's just storing the words linearly in the order they are imported. I'm not sure how this affects the probability component since words have multiple instances. though maybe he's taking that into account when he retrieves the word rather than when he puts it in. It also looks like he's using a graph database for something? know they are trendy, but seems weird.. Maybe it's just easier to work with than other methods like my stupid hacky arrays.
The sentences don't make a lot of sense, but that is as expected. This concept is super basic, and the output intended to be more inspirational than usable verbatim. and hell, most song lyrics don't make a lot of sense anyway.
Here's an example of some of the things that come out based on a corpus of lyrics from the band hum:
my eye to see far to spare and she miss the train.
differ from her dream with absent eye.
upright let me to come back now.
do care what they choos.
dream with seven nine and think about the end.
I guess the biggest issue I have is that he used the NLTK routine for "stemming" the words, hence some of them missing their suffixes. If this were intended to be generating grammar from scratch, that would matter, but I'm just going off of probability, so it doesn't. He said he'll pull it out in the next rev, and I think the usefulness of output will go up a lot.
I only briefly looked over the code. It's not how I would have done it / want it to be in some ways, but I guess in terms of utility, it still gets me farther than I was. For example, he's just storing the words linearly in the order they are imported. I'm not sure how this affects the probability component since words have multiple instances. though maybe he's taking that into account when he retrieves the word rather than when he puts it in. It also looks like he's using a graph database for something? know they are trendy, but seems weird.. Maybe it's just easier to work with than other methods like my stupid hacky arrays.
The sentences don't make a lot of sense, but that is as expected. This concept is super basic, and the output intended to be more inspirational than usable verbatim. and hell, most song lyrics don't make a lot of sense anyway.
