Machine Learning

I-Fingerprint yesibili-yesibili: ngaphakathi kwengoma esheshayo ye-Shazam


Lokhu okuthunyelwe kuyaqhubeka ngemuva kwempompo, uchungechunge oluhlola imishini efihlekile yezobuchwepheshe yansuku zonke – kusuka ku-Uber to Spotify ukusesha izinjini. Ngizothambekela ngaphansi kwe-hood ukudiliza amasistimu abumba umhlaba wakho wedijithali.

Ubudlelwano bokuqala obunokulalela umculo buqale ngo-6, bajikeleza ngama-albhamu esidlali segumbi le-ONKYO 6-disc player. Ikati izithuthi, Inkatho Impindanda, Isikhathi. Kwakukhona njalo ngengoma eyodwa engaqhubeka nokuqala kabusha, yize ngangingalazi igama laso. Iminyaka eyi-10 kuqhubeke, izikhathi zengoma ibuyele kwimemori. Ngiseshe izinkundla, 'i-saxophone melody ''Ingoma yeVintage mayelana nezindunduma zesihlabathi', Ufuna iminyaka ngaphandle kwempumelelo. Lapho-ke, ngolunye usuku eyunivesithi, ngangisegumbini lomngane wami uPegled Pegled lapho edlala ngalo:

https:/www.youtube.com/watch?v=wzct7abuu9y

Lokho kusesha isikhathi eside kwangifundisa ukuthi kubaluleke kangakanani ukuthola umculo owuthandayo.


Ngaphambi kokusakazwa kanye nabalalisa abahlakaniphile, ukutholwa komculo kuncike kwimemori, inhlanhla, noma umngani ngokunambitheka okuhle komculo. Ukuthi i-chorus eyodwa ebabazekayo ingahle ilahlekelwe yi-ether.

Kwabe sekufika isimangaliso somculo womculo.

Imizuzwana embalwa yomsindo. Cindezela inkinobho. Negama esikrinini sakho.

UShazam wenza umculo wabonakala.

Imvelaphi: 2580

UShazamu wethule ngonyaka ka-2002, isikhathi eside ngaphambi kokuthi izinhlelo zokusebenza zaziyinto. Ngaleso sikhathi kwasebenza kanjena:

Uzoshayela 2580 # kuselula yakho (UK kuphela).
Bamba ifoni yakho kuze kube yisipika.
… Linda uthule …
Futhi wamukele a I-SMS Ukutshela igama lengoma.

Kwakuzwakala njengomlingo. Iqembu Lokusungulwa, uChris Barton, uPhilip Englelbrecht, u-Avery Wang, noDhiraj Mukelee, bachitha iminyaka beyakha inkohliso leyo nkohliso.

Ukwakha i-database yayo yokuqala, i-Shazam yaqasha abasebenzi abangama-30 abancane ukuthi basebenze amashifu amahora ayi-18, balayishe ngesandla ama-100,000 ama-CD kumakhompyutha futhi basebenzise isoftware yangokwezifiso. Ngoba ama-CD awaqukethe i-metadata bekufanele athayiphe amagama ezingoma ngesandla, ebhekisa kumkhono we-CD, ekugcineni adale iminwe yeminwe yezigidi zenkampani yezigidi – inqubo ebuhlungu eyathatha izinyanga.

Ngesikhathi ngaphambi kwama-Smartphones noma izinhlelo zokusebenza, lapho ama-Nokia's kanye ne-blackberry akwazanga ukuphatha ukucutshungulwa noma ukukhumbula izimfuno, i-shazam kufanele ihlale iphila isikhathi eside ngokwanele ubuchwepheshe ukuze bathole umbono wabo. Lesi kwakuyisifundo ngesikhathi semakethe.

Lokhu okuthunyelwe kumayelana nokuthi kwenzekani okwamanje ngemuva kompompi kanye nesihloko, ukucubungula kwesiginali, ukucubungula, ukufana okukhomba, nokufana okuvumela i-Shazam ukuzwa ukuthi yini ongakwazi ukukuzwa.


I-Algorithm: Ukuvunyelwa kweminwe okulalelwayo

Ngo-2003, i-Shazam Counder Avery Wang yashicilela i-BluePrint ye-algorithm eseyinamandla uhlelo lokusebenza namuhla. Umbono wephepha eliphakathi: Uma abantu bengakuqonda umculo ngo ukucindezela Izendlalelo zomsindo, umshini ungakwenza futhi.

Ake sihambe ngendlela u-Shazam ephuka ngayo azwakala emshinini ongawabona khona manjalo.

1. Ukuthwebula isampula yomsindo

Kuqala ngompompi.

Uma ushaya inkinobho ye-Shazam, uhlelo lokusebenza luqopha i-Snippet yesibili yomsindo okuzungezile. Lokhu kudala ngokwanele ukukhomba izingoma eziningi, yize sonke silinde imizuzu ebambe izingcingo zethu emoyeni (noma ukucasha emaphaketheni ethu) nge-ID.

Kepha uShazamu akagcini ngokuqoshwa. Esikhundleni salokho, kunciphisa okuthile okuncane kakhulu futhi kube ngobuchule: a umnyatheliso weminwe.

2. Ukudala i-spectrogram

Ngaphambi kokuthi u-Shazam angabona ingoma, kudingeka aqonde ukuthi yini Imvamisa inomsindo futhi lapho zenzeka. Ukuze wenze lokhu, usebenzisa ithuluzi lezibalo elibizwa ngokuthi i-Fast Fourier Transform (FFT).

Le khasi I-FFT Iphula isinali elalelwayo kumafrikhwensi alo, embula ukuthi imaphi amanothi noma amathoni akha umsindo nganoma yisiphi isikhathi.

Okwenza Kubalulekile: Ama-waveforms antekenteke, azwela umsindo, izinguquko zokuphikisana, kanye nokucindezelwa kwedivayisi. Kepha ubudlelwane bokuvama ngokuhamba kwesikhathi buhlala buzinze. Lelo yigolide.

Uma ufunda izibalo e-Uni, ungakhumbula ubunzima bokufunda inqubo eguqukayo eyi-discrete.I-Fast Fourier Transform (FFT) Inguqulo esebenza kahle evumela ukuthi sibole isiginali esiyinkimbinkimbi zibe yizingxenye zayo zemvamisa, njengokuzwa wonke amanothi.

Umculo awulona i-static. Amanothi kanye ne-harmonics kuguquka ngokuhamba kwesikhathi. Ngakho-ke uShazam akagcini nje ngokuqhuba i-FFT kanye, kuyayigijima kaninginingi ngaphezulu kwamafasitela amancane, agcwele ngaphezulu kwesiginali. Le nqubo yaziwa ngokuthi Isikhathi esifushane esifushane (STFT) futhi yakha isisekelo se i-spectrogram.

Isithombe nguMlobi: Ukuguqulwa okusheshayo okune-Fourier okubonwayo

Umphumela i-spectrogram uguquko lomsindo kusuka ku Isizinda sesikhathi se-Amplitude (Waveform) ku Isizinda sesikhathi sokuvama.

Cabanga ngalokhu njengokuguqula i-wavelform engcolile yomsindo ibe yi-heatmap yomculo.
Esikhundleni sokubonisa ukuthi umsindo unjani, imibukiso ye-spectrogram Imiphi imvamisa bakhona Ngaziphi izikhathi.

Image by Umbhali: Ukubonwa kwenguquko kusuka ku-waveform kuya ku-spectrogram usebenzisa i-FFT

I-spectrogram ihambisa ukuhlaziya kusuka ku Isizinda sesikhathi se-Amplitude kwa- Isizinda sesikhathi sokuvama. Ibonisa isikhathi kwi-axis evundlile, imvamisa eksishini mpo, futhi isebenzisa ukukhanya ukukhombisa i-amplitude (noma ivolumu) imvamisa ngayinye ngesikhathi ngasinye. Lokhu kukuvumela ukuthi ungaboni ukuthi yimiphi imvamisa, kodwa futhi nokuthi ubukhulu bawo buvela kanjani, okwenza bukwazi ukubona amaphethini, imicimbi yesikhashana, noma izinguquko kusiginali ezingabonakali ku-waveform yesikhathi esijwayelekile.

Ama-spectrograms asetshenziswa kabanzi emikhakheni efana nokuhlaziywa komsindo, ukucubungula izinkulumo, i-seismology, nomculo, ukuhlinzeka ngethuluzi elinamandla lokuqonda izici zesikhashana nezibukeli zezimpawu.

3. Kusuka ku-spectrogram kuya kumephu ye-constellation

Ama-spectrograms aminyene futhi aqukethe idatha eningi ukuqhathanisa izigidi zezingoma. I-Shazam ihlunga imvamisa ephansi-ubukhulu, ishiye iziqongo eziphakeme kakhulu.

Lokhu kudala imephu yenhlangano, isaphulelo esibonakalayo samaza wokusukuma ngokuhamba kwesikhathi, afana neshidi lomculo, yize kungikhumbuza ibhokisi lomculo lemishini.

Isithombe nguMlobi: Ukubona ngeso inguquko kwimephu ye-constellation

4. Ukwakha iminwe yomsindo

Manje kuza umlingo, ukuguqula amaphuzu abe yisiginesha.

I-Shazam ithatha indawo ngayinye ye-anchor (i-Peak evelele) futhi iyibhangqa ngeziqongo ezihlosiwe ngewindows lesikhathi esincane phambili – ukwakha ukuxhumana okuhlanganisa umehluko wemvamisa nokwesikhathi.

Ngayinye yalezi iba yi-hashi tuple:

(i-anchor_frequency, okuhlosiwe_frequency, isikhathi_delta)

Isithombe nguMlobi: Inqubo yokuGukunzela Hash

Yini i-hashi?

I-hash ukukhishwa komsebenzi wezibalo, obizwa ngokuthi umsebenzi we-hashi, oguqula idatha yokufaka ibe umucu wobude obuhleliwe kanye / noma izinhlamvu. Kuyindlela yokuguqula idatha eyinkimbinkimbi ibe isikhombi esifushane, esiyingqayizivele.

I-Hashing isetshenziswa kabanzi kwisayensi yekhompyutha kanye ne-cryptography, ikakhulukazi imisebenzi efana nokubuka idatha, ukuqinisekiswa, kanye nokukhomba.

Isithombe nguMlobi: bhekisa kulo mthombo uqonde ukushona

Nge-shazam, okujwayelekile I-Hash yi-32 bits eside, futhi amadla ahlelwe kanjena:

  • Amabhithi ayi-10 ngokuvama kwe-anchor
  • Amabhithi ayi-10 ngokuvama okuhlosiwe
  • Amabhithi ayi-12 Okwesikhathi Delta phakathi kwabo
Isithombe nguMlobi: Ukubonwa kwe-having eyisibonelo kusuka phezulu

Lo munwe omncane uthwebula ubudlelwano phakathi kweziqongo ezimbili zomsindo nokuthi ziqhele kangakanani ngesikhathi, futhi ziqinile ukuze zikhombe ingoma nezincane ngokwanele ukuba zidlulise masinyane, ngisho nokuxhumeka komkhawulokumbili okuphansi.

I-5. Ukuqhathanisa ngokumelene database

Lapho uShazam edala umunwe ovela ku-Snippet yakho, udinga ukuthola ngokushesha umdlalo ku-database yawo equkethe izigidi zezingoma.

Yize uShazamu akanandaba ukuthi iculo lakho liqhamuka kuphi – i-intro, ivesi, i-chorus, ibhuloho – alinandaba, libheka isikhathi esilinganiselwe phakathi kwama-hashi. Lokhu kwenza uhlelo luqine isikhathi eside kulalelwayo lokufaka.

Image by Umbhali: ukubona ngeso lengqondo ama-hashes ahambelana nengoma yedatha

UShazamu uqhathanisa ama-hashes akho okuqoshwa ngokumelene nemininingwane yawo futhi akhomba ingoma enenombolo ephezulu kakhulu yemidlalo, iminwe enemigqa engcono kakhulu ngesampula yakho, noma ngabe akuyona umdlalo oqondile ngenxa yomsindo wangemuva.

Iseshwa kanjani ngokushesha okukhulu

Ukwenza lo mbani-ngokushesha, i-shazam isebenzisa a hashmap, Isakhiwo sedatha esivumela ukubheka okusondele ngokushesha.

I-hashmap ingathola umdlalo ku-O (1) isikhathi, lokho kusho isikhathi sokubheka sihlala njalo, noma ngabe kunezigidi zokufakwa.

Ngokuphambene nalokho, inkomba ehleliwe (efana ne-B-Tree kudiski) ithathe isikhathi se-O (log n) isikhathi, esikhula kancane njengoba i-database ikhula.

Le esele yesikhathi nesikhathi esiyinkimbinkimbi saziwa ngokuthi yi-Big O Notation, umbono angilungiselelwe ukukhathazwa ukufundisa. Sicela ubheke usosayensi wekhompyutha.

6. Ukukala uhlelo

Ukuze ulondoloze le jubane esikalini somhlaba wonke, i-Shazam yenza okungaphezu kokusebenzisa izakhiwo zedatha esheshayo, yandisa ukuthi idatha ihlala kuphi futhi kuphi:

  • Shards I-database – ukuyihlukanisa nge-Time Range, i-Hash Prefix, noma iJografi
  • Igcina ama-shard ashisayo kwimemori (i-RAM) yokufinyelela okusheshayo
  • I-Offleats Collelle Idatha ediski, ehamba kancane kodwa eshibhile yokugcina
  • Isabalalisa uhlelo ngesifunda (isib. US East, Europe, e-Asia) Ngakho-ke ukuqashelwa kuyashesha noma ngabe ukuphi

Lo mklamo uyasekela 23,000+ ukuqashelwa ngomzuzungisho nasekulinganisweni komhlaba.


Umthelela & Izicelo Zesikhathi Esizayo

Uhlelo lokusebenza olusobala ukutholwa komculo ocingweni lwakho, kepha kukhona olunye uhlelo lokusebenza olukhulu lwenqubo kaShazamu.

I-Shazam iyasiza Ukuqonda Kwemakethe. Njalo uma kutheke umaki womsebenzisi ingoma, u-Shazam aqoqa ngokungaziwa, i-metadata ye-geo-temporel (lapho, khona, futhi kuvame kangakanani ukuthi ingoma ingoma i-ID.)

Amalebula, abaculi, kanye nabagqugquzeli bakusebenzisa lokhu:

  • Amathrekhi we-Spot Breakout ngaphambi kokushaya amashadi.
  • Khomba izitayela zesifunda (i-remix yokuthola umkhondo eTokyo ngaphambi kweLa).
  • Qondisa ukuthengisa imali ngokususelwa ekuheheni kwe-organic.

Ngokungafani ne-Spotify, esebenzisa indlela yokulalela yomsebenzisi yezincomo zokuhlaziywa, i-Shazam inikezela ngedatha yesikhathi sangempela ngezingoma abantu abazikhomba ngenkuthalo, banikeze imboni yezomculo kuqala ekuboneni kwamathrendi avelayo namathrekhi athandwayo.

Yikuphi ukuzwa okuzwayo ngaphambi kokwenza
Isayensi yedatha yokuncoma komculomedium.com

NgoDisemba 2017, i-apula uthenge i-shazam for a kubikiwe $ 400 million. Kubikwa ukuthi i-Apple isebenzisa idatha ye-shazam ukwengeza injini yokuncoma ye-Apple Music, futhi ukuqopha amalebula manje aqapha ukuthambekela kwe-Shazam njengoba babevame ukuqapha imisakazo spins.

Izithombe nguRachel Coyne ku-Undercwas

Esikhathini esizayo, kulindeleke ukuthi kube nokuziphendukela kwemvelo ezindaweni ezinjengokuthi:

  • I-Visual Shazam: Usuvele uphululwe, akhombe ikhamera yakho entweni noma yobuciko ukukhomba, iwusizo ku-Nex Agementented Reality Metual.
  • Imodi yekhonsathi: Khomba izingoma bukhoma phakathi kwama-gigs futhi uvumelanise ohlwini lokusetha lwesikhathi sangempela.
  • Izitayela ze-Hyper-Local: Ingaphezulu Yenzelwe Ukuthambekela 'Kulomgwaqo' noma 'kule ndawo', ukunwebeka ukunambitheka komculo okwabiwe kulo mphakathi.
  • Ukuhlanganiswa kwe-AI kwe-AI: PAma-Snippets audio air nge-lyric Generation, Remix iziphakamiso, noma ukuhambisana okubukwayo.

I-OUTHO: I-algorithm ekhuthazayo

Emhlabeni wezitaki ze-Tech eguqukayo, akuvamile ukuthi i-algorithm ihlale ifanele iminyaka engaphezu kwengu-20.

Kepha indlela yeminwe ye-shazam ayikaze ikhule, iyavela, yavela, futhi ibe yi-blueprint yezinhlelo zokubona umsindo kuyo yonke imboni.

Umlingo akuyona nje ukuthi i-shazam ingaqamba igama. Kukwenza kanjani, ukuguqula umsindo ongcolile ube yizibalo ezinhle, futhi ukukwenza ngokuthembekile, khona manjalo, nasemhlabeni jikelele.

Ngakho-ke ngokuzayo lapho use-liper, ibha yedoti ebambe ifoni yakho kuze kube yilapho uSomlomo edlala ULola Young 'messy' Vele ukhumbule: Ngemuva kwalolo mpompi yisitaki esihle sokucutshungulwa kwesiginali, ukugeza, nokusesha, kwahlelwe kahle kakhulu ukuthi kwakudingeka kushintshe.

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button