{"slug": "i-used-dettato-to-obtain-bass-tabs-from-youtube-video", "title": "I used dettato to obtain bass tabs from YouTube video", "summary": "A blogger used dettato, a newly released audio-to-MIDI neural network trained on public domain music, to transcribe the 7/8-time bass line of Angine de Poitrine's \"Fabienk | Part II\" from a YouTube MP3 and published the resulting bass tab and MIDI for download. After building a fork with uv tooling, Apple Silicon MPS support and a GUI, the author had OpenAI's Codex (model Terra) isolate the repeating bass motif from the full MIDI and convert it to tab and musical score, each in a single attempt. The author published the tab and a browser-based practice player, noting the riff is simple but hard to play at the band's tempo.", "body_md": "# Angine de Poitrine bass tab and how I got it using AIs\n\nHere I present for download and practice bass tab of the Angine de Poitrine – Fabienk | Part II. Below I describe how I utilized AI (dettato) to get it.\n\nYou can download the midi here:\n\nbass_motif_7-4_original_variation-best-Kontrabas,_Bass_motif343 B[Download](https://www.slowchronicle.com/media/cdb70844-824a-4a2b-b5eb-2d87aea1f5e2?download=1)\n\nAnd practice live in browser here. Change the tempo or play just the drums to fit your needs.\n\n### Practice the tab\n\n## \n\nFew days ago, new audio-to-midi model was released. Its called [dettato.](https://github.com/kobimusic/dettato) Its not and LLM. Its different kind of neural network trained on one and only task - to convert audible music into midi notation. That is into sequence of individual notes.\n\nWhat is cool about this specific one, is that it was trained on public domain music, that it provides good results, and that it extracts several tracks of the original audio (ensemble, guitar, singing, bass, drums, literally anything). I decided to test it out on  song Fabienk of the rising new band [Angine de Poitrine](https://www.youtube.com/watch?v=pRgHYWOtqqc).  Its really unique band in that it uses untraditional rhythms (this song is in 7/8) and microtonals (that means tones in between semitones - they have special guitar for that with much more frets than is usual).\n\nI especially liked the bass line in the second section of the song. There is not much microtones (or can be easily avoided) but I was unable to note down the individual notes from the musicians fast playing.\n\nFantastic opportunity to try out the new dettato model, occured to me! First I made my own fork, with uv tooling, MPS (Mac Silicone) support and simple to use GUI.\n\nThen I converted MP3 extracted from their youtube video, into midi. Once i opened it in garageband and replayed from midi, it surprised me how well the music was transcribed. Guitar, bass, drums.\n\nAll-tracks converted midi53 KB[Download](https://www.slowchronicle.com/media/0e222b70-f5b4-449e-866d-49239fe709fa?download=1)\n\nBut I wanted only the bass. Thus I instructed AI (codex) to read this midi, and look at the bass part in the second half of the song. And to find the repeating motiff and give it to me in separate midi file. Ha! The codex (I think I used model Terra) one-shotted it. It was perfect.\n\nThen I just asked it to convert it to me into bass tabs, and musical score, which was also one-shotted, and then I was able to start learning it on my bass guitar.\n\nTurns out the melody is really not that difficult. Its the speed. I have not matched the performance of the Angine de Poitrine themselves and probably will not match in coming days.\n\nYet still I decided to publish this to youtube and to public. This bands music is getting extremely popular in last months and I guess people will be happy to find the tabs to learn to play it by themselves.\n\nAI proves to be really useful in many areas of life. Be it coding, or extracting repeating patterns from specific track of midi partiture. The more advanced they are, the more one-shot it all becomes. Thats really great.\n\nBut its also worh noting, that non-LLM models are also still being developed and have their great use. And dettato is excelent example of that.\n\nSo grab your bass and learn this riff. In at least the Angine de Poitrine's tempo!", "url": "https://wpnews.pro/news/i-used-dettato-to-obtain-bass-tabs-from-youtube-video", "canonical_source": "https://www.slowchronicle.com/posts/angine-de-poitrine-bass-tab", "published_at": "2026-11-22 20:41:55+00:00", "updated_at": "2026-11-22 21:11:48.188410+00:00", "lang": "en", "topics": ["machine-learning", "neural-networks", "ai-tools", "ai-products"], "entities": ["dettato", "Angine de Poitrine", "Fabienk | Part II", "Codex", "OpenAI", "GarageBand", "YouTube"], "also_reported_by": [], "takeaway": "A task-specific audio-to-MIDI model (dettato) transcribed guitar, bass and drums from an MP3, and Codex then isolated the repeating bass motif and produced tab and score in one shot each.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/i-used-dettato-to-obtain-bass-tabs-from-youtube-video", "markdown": "https://wpnews.pro/news/i-used-dettato-to-obtain-bass-tabs-from-youtube-video.md", "text": "https://wpnews.pro/news/i-used-dettato-to-obtain-bass-tabs-from-youtube-video.txt", "jsonld": "https://wpnews.pro/news/i-used-dettato-to-obtain-bass-tabs-from-youtube-video.jsonld"}}