# Project Progress - **Title**: Sound Classification on rap vocals and speech - **Subtitle**: Using `Demucs` & `yt-dlp` for Music Analysis *Vio Chung* *03/05* ## Outline - Collect Data (finished) - Preprocess Data (just strated) > Build Pipline and Model Training Fine Tune Model Test Model Evaluate Model Preformance ### The Data - rap & speech 1. Rap is a rather new music style, and it's very culture oriented. So I try to make sure that the variations are equally covered in my data collection. :::info Arranged by genres: :::spoiler 1. Old School Rap * Era: Late 1970s to mid-1980s * Characteristics: Characterized by simple rhymes, relatively basic beats, and an emphasis on party-oriented themes. Pioneering artists include Grandmaster Flash and the Furious Five, Run-D.M.C., and LL Cool J. * "The Message" by Grandmaster Flash & The Furious Five * "Rapper's Delight" by The Sugarhill Gang * "Planet Rock" by Afrika Bambaataa & The Soulsonic Force * "Paid in Full" by Eric B. & Rakim * "Ladies First" by Queen Latifah featuring Monie Love (female)
 2. Gangsta Rap * Era: Mid-1980s onwards * Characteristics: Focuses on the violent lifestyles and impoverished conditions of inner-city life. Gangsta rap is known for its explicit lyrics and is represented by artists like N.W.A, Ice-T, and later by Tupac Shakur and Notorious B.I.G. * "Fight the Power" by Public Enemy * "Alright" by Kendrick Lamar * "The Message" by Nas * "Umi Says" by Mos Def * "Lost Ones" by Lauryn Hill (女)
 3. Conscious Rap * Era: Late 1980s onwards * Characteristics: Known for its focus on social issues, consciousness, and the need for political change. It often tackles topics like racism, poverty, and injustice. Prominent artists include Public Enemy, KRS-One, and more recently, Kendrick Lamar. * "Fight the Power" by Public Enemy * "Alright" by Kendrick Lamar * "The Message" by Nas * “Ergonomics” by Ill Conscious * "Lost Ones" by Lauryn Hill (女)
 4. Trap * Era: Early 2000s onwards * Characteristics: Originated in the Southern United States, trap music is known for its lyrical content that often focuses on the "trap" lifestyle of selling drugs and its hard-hitting beats. Artists like T.I., Gucci Mane, and Migos are well-known trap musicians. * "Rubber Band Man" by T.I. * "Lemonade" by Gucci Mane * "Bad and Boujee" by Migos * "Bodak Yellow" by Cardi B (女) * "SICKO MODE" by Travis Scott
 5. Drill * Era: Early 2010s onwards * Characteristics: A subgenre of rap that originated in Chicago, drill music is noted for its dark, grim, and violent lyrical content, reflecting the harsh realities of urban life. Chief Keef and King Von are notable artists within this genre. * "I Don't Like" by Chief Keef * "Welcome to the Party" by Pop Smoke * "Crazy Story" by King Von * "No Suburban" by Sheff G
 6. Mumble Rap * Era: Mid-2010s onwards * Characteristics: Characterized by a perceived lack of enunciation and lyrical complexity, mumble rap focuses more on melody and vibe. Artists such as Future, Lil Uzi Vert, and Playboi Carti are often associated with this style. * "XO TOUR Llif3" by Lil Uzi Vert * "Gucci Gang" by Lil Pump * "Magnolia" by Playboi Carti * "m.A.A.d city" by Kendrick Lamar * “Comic Sans” by AUDREY NUNA (女) * “damn Right” by AUDREY NUNA (女)
 7. East Coast Rap * Characteristics: Known for its complex lyrics, intricate rhymes, and often jazzy beats, East Coast rap originated from New York City and includes artists like Jay-Z, Nas, and the Wu-Tang Clan. * "Juicy" by The Notorious B.I.G. * "N.Y. State of Mind" by Nas * "C.R.E.A.M." by Wu-Tang Clan * "Get Ur Freak On" by Missy Elliott (女） * "Mass Appeal" by Gang Starr
 8. West Coast Rap * Characteristics: Features funk-influenced beats and a laid-back vibe, reflecting the lifestyle of the West Coast, particularly California. Dr. Dre, Snoop Dogg, and Tupac Shakur are iconic figures in this genre. * "Nuthin' but a 'G' Thang" by Dr. Dre featuring Snoop Dogg * "California Love" by 2Pac featuring Dr. Dre * "It Was A Good Day" by Ice Cube * "The Next Episode" by Dr. Dre featuring Snoop Dogg, Kurupt, Nate Dogg * "Fancy" by Iggy Azalea featuring Charli XCX (女) 
 9. Southern Rap * Characteristics: Encompasses a range of styles but is generally known for its distinct sound that includes chopped and screwed techniques, heavy use of bass, and a slower tempo. OutKast, Scarface, and Lil Wayne are significant artists from this genre. * "International Players Anthem (I Choose You)" by UGK featuring OutKast * "Ms. Jackson" by OutKast * "Stay Fly" by Three 6 Mafia * "Savage" by Megan Thee Stallion （女) * "A Milli" by Lil Wayne
 10. Alternative Rap * Characteristics: Stands out by incorporating elements from various genres, including funk, rock, and jazz, often with a focus on experimental sounds and socially conscious lyrics. Artists like A Tribe Called Quest, De La Soul, and Tyler, The Creator exemplify this genre. * "Can I Kick It?" by A Tribe Called Quest * "Me Myself and I" by De La Soul * "Rebirth of Slick (Cool Like Dat)" by Digable Planets * "Man I Used to Be" by k-os * "Paper Planes" by M.I.A. (女)
 11. Grime * Era: Early 2000s onwards * Characteristics: A genre that emerged in London, characterized by its rapid breakbeats, electronic sounds, and aggressive lyrics. Artists such as Dizzee Rascal, Stormzy, and Skepta are prominent figures in grime. * "Shutdown" by Skepta * “Gas Me Up” by Skepta * "Did You See" by J Hus * “Juicy Freestyle” by Bryn (女) * “Rumours” by Ivorian Doll (女) ::: >Found that the amount of female rap artists are lesser than male rap artists. >I try to balance the data by adding more songs from other female artists just try not to make a big difference between gender. 2. As for speech data, I find that most videos that have speech in it also have bgm. >-> good thing for data consistency :::warning :chart: **Overall Video collection:** 120 rap videos ≈ 11 hours 83 speech videos ≈ 13 hours ::: ### Tools & Technologies Used at this moment - Command line code `yt-dlp` for gathering data from Youtube. - Demucs for rap vocal separation. ## yt-dlp - Data Gathering - The command line code used to grab data from YouTube. ``` yt-dlp --output "/Users/vio/yt-dlp/%(title)s.%(ext)s" --write-info-json --extract-audio --audio-format mp3 --audio-quality 320K “SomeYoutubeLink" ``` - Online JSON viewer to inspect the files ![Screenshot 2024-03-05 at 3.38.33 PM](https://hackmd.io/_uploads/ryFCfWS6T.png) ## Demucs - Separating Vocals from Tracks - a music source separation model. >separating drums, bass, and vocals from the rest of the accompaniment. >It's a python library. - code: >![Screenshot 2024-03-05 at 3.27.38 PM](https://hackmd.io/_uploads/HJPrx-Saa.png) - ==Switch to file demonstration==. ### References - [Demucs](https://github.com/facebookresearch/demucs) - [yt-dlp Github](https://github.com/yt-dlp/yt-dlp)