Using GPT instead of Music Pikard to return the year and genre of my music database, hoping for more clean/accurate results.
But now, how to integrate that data back into Lexicon > then Serato?
That’s cool but I wouldn’t count on more accurate results. ChatGPT is known for making up things.
There is no easy way to get those results back into Lexicon currently. It is possible through the API but you’d have to write a script.
1 Like
I’ll get GPT to write that API code!
Agree with your sentiment, but Pikard is also not working ideally
Especially with new and mashup remixes, GPT is picking up the details correct
I was checking with you if Lexicon had a similar import feature, as it does with CSV export
1 Like
Not at the moment, generally Lexicon does that for you automatically like with finding missing tags. The API is there for techy folks who want more.
Not saying that ChatGPT isn’t super interesting and maybe it can help in Lexicon, who knows, further investigation needed 
I am absolutely thrilled by the capabilities of GPT! It is incredible to see how it can detect even the release dates of “mashup and DJ edit” tracks from record pools.
This implementation would be immensely beneficial to me as the current ‘Find Tag/Album art’ feature has not been able to provide me with accurate data. With this feature, I can better plan and organize my music folders without any hassle.
I am confident that this feature will be helpful to countless other users like me who have a vast music collection. In fact, I would be more than happy to upgrade my subscription to take advantage of this feature. This would enable me to quickly and accurately organize my 30,000 tracks without having to deal with the learning curve and poor results that pikard offers.
I decided to test it out with the classic country hit, “Coward of the County” by Kenny Rodgers, and I was amazed by the results obtained from both LEXICON and GPT. (note GPT it avoids confusion with additional META data from MIXED IN KEY, which exists as the number “-3” in the song title.
2 Likes
In my last example, you will see how LEXICON decides that the 1979 Country classic, is categorised instead as a 2009 Dance Electro Pop song.
FYI - here is a link to the actual song
I suspect that LEXICON is reporting the latest album, which the song was released on - same as PIKARD et al.
But how, Dance, Electro Pop!?
It’s definitely a valuable tool and certainly something that would be good in the Find Tags & Album Art tool. The UI would have to change I think, so you can preview the results and save them if they look good
1 Like
i actally use chat gpt at gigs when I’m asked to play something (a genre) I’m not familiar with, and ask for recommended songs with bpm, artist- title and list in ascending order. its works about 70% of the time … so AI nothing is perfect!
2 Likes
Yes, the same here!
It works to overcome mental blocks and as a source of inspiration - not perfect, but gets you past a difficult moment nonetheless!
Another tip, is to ask GPT To <present in a simple table, without the “”>
Then its a copy/paste job into your DJ software!
I’ve been thinking about ChatGPT lately and how it can help Lexicon, I think it could be rather useful.
Let me know any way you use it in regards to DJing, that really helps me understand how we could utility it in Lexicon
2 Likes
I know this is an old thread, but I’m working on something similar to this in a Python script. I initially tried just sending the track to ChatGPT but the problem is that none of the models that actually can search the web are accessible via API. The models you can script will result in hallucinations. So I’m making it a bit more complicated in order to search the web myself, then give the info to GPT to choose the best match. The flow will look something like this:
- Get track data from Lexicon API
- Search for track online using APIs and scraping (iTunes/AM, Spotify, Beatport, Bandcamp, SoundCloud, MusicBrainz, YouTube). Get as many fields as possible from each.
- Normalize the results to extract Base Title Name, Mix, Artists, Featured Artists, Remixer.
- Do some basic comparisons to the input track to get rid of the garbage results. I.e. title, artist, duration, etc must be somewhat similar.
- Use the Song.Link/Odesli API for each result to grab IDs for the same track on other services. Use this info to group the results into buckets of the same track.
- Send these results to a basic ChatGPT model (gpt-4o-mini) to choose the group that best matches the input. Have it return info collapsed from the chosen group, and in the format I want things to be organized in.
- Update track using Lexicon API.
I’ll definitely be sharing all this on github when I’m done, maybe some of it will be useful.
2 Likes
For what it’s worth and I know it’s a manual exercise for me but " genius.com " works for me and has been the most accurate 
I wish there was an easier and more accurate way.
1 Like