Up tókànItọsọna atẹle
SI-SDR fun Pipin Ohun (Audio Source Separation)
Ohùn AI
Audio AI Itọsọna
Orisun-Filter Vocoding ati WORLD
Modeli source-filter pin si i ṣiṣẹda ohun (speech excitation) kuro ni iṣeto oníṣọ́ (acoustic shaping) ti ọ̀nà ohun (vocal tract).
Gbọ si itọsọna yiiAI-ṣe afihan · ~2 min
Lori iwe yi2 min ka
Akopọ
WORLD jẹ́ vocoder ti o ṣe iṣiro ati dahun ohun (resynthesizes speech) nipa lilo iye igbohunsafẹfẹ ipilẹ (fundamental frequency), awọ-ara spectral (spectral envelope), ati aperiodicity. A le ṣe atunṣe awọn paramita wọnyi lọtọ ninu ilana iṣakoso ami (signal-processing workflow), botilẹjẹpe atunṣe ko jẹrisi pe aṣoju oluwo (speaker identity) tabi dandan (naturalness) yoo wa kanna.
Jin Dive
Ninu modeli source-filter, source n ṣe afihan iṣiṣẹ awọn ẹyẹ ohun (vocal-fold excitation) fun ohun ti o ni ohun (voiced) tabi iṣiṣẹ ohun-ọna (noise-like excitation) fun ohun ti ko ni ohun (unvoiced), nigba ti filter n ṣe afihan awọn iduro (resonances) ti ọ̀nà ohun (vocal tract). WORLD jẹ́ vocoder ti a ṣe apẹrẹ fun iṣẹda ohun (speech synthesis) ti o ni didara giga ati awọn iṣẹ akoko gidi (real-time applications). O n ṣe iṣiro F0, awọ-ara spectral, ati aperiodicity, lẹhinna o n lo awọn ohun-ini wọnyi lati ṣẹda ohun. Awọn algoridimu rẹ pẹlu DIO tabi Harvest fun F0, CheapTrick fun awọ-ara spectral, ati D4C fun aperiodicity. Gbigbe awọn paramita lọtọ n fun laaye atunṣe ti o ni iṣakoso, bii yiyan F0 lakoko ti o n gbe awọ-ara spectral (envelope stream) duro, ṣugbọn awọn aṣiṣe iṣẹda (synthesis artifacts) ati iyipada ti a le ri (perceptual changes) le ṣẹlẹ.
Real-World imuse
Awọn irinṣẹ iyipada ohun ti o yipada ipolowo agbọrọsọ ati timbre lakoko ti o jẹ ki ọrọ ni oye
Awọn olupilẹṣẹ ohun orin (gẹgẹbi ilolupo eda UTAU/NNSVS) ti o tun ṣe awọn akọsilẹ ni awọn ipolowo tuntun
Awọn ọna ṣiṣe ọrọ-si-ọrọ parametric ti o ṣe agbekalẹ F0, spectral, ati awọn ṣiṣan aperiodicity ṣaaju ki o to fokodi
Awọn ipilẹ ti iwadii ọrọ fun iyipada ipolowo, gigun akoko, ati ṣiṣatunṣe prosody laisi atunṣeto
Ojo iwaju ti Orisun-Filter Vocoding ati WORLD
Awọn eto ohun ti o jẹ hybrid le darapọ awọn alaye F0 ati spectral ti o rọrun lati loye pẹlu iṣẹda waveform ti o da lori neural. WORLD tun wulo fun iṣiro iyara, iṣẹda ohun ti o ni iṣakoso, ati afiwe iwadii, nigba ti awọn ọna neural tuntun ma n ṣe iwadii fun dandan (perceptual naturalness). Awọn iṣẹlẹlẹ yẹ ki o ṣe idanwo awọn aṣiṣe (artifacts), oye (intelligibility), ati ifarakanra oluwo (speaker similarity) lẹhin ti a ba yi awọn paramita pada.
Fifi o sinu iwa
Kini idi ti o ṣe pataki
- Wiwọle ati arọwọto. O ṣe ilọsiwaju iraye si nipasẹ transcription, alaye, ati awọn atọkun ohun.
- Iye owo ati isuna. Awọn ẹgbẹ Media le firanṣẹ ohun didan yiyara pẹlu awọn isuna-owo kekere.
- Iyara ati iwọn. Awọn ọna ṣiṣe ti nkọju si alabara le ṣe ilana awọn ibaraẹnisọrọ sisọ ni iwọn nla.
Ṣọra fun
- ilokulo ohun ati awọn ewu afarawe ṣe pọ si nigbati igbanilaaye ba sonu.
- Yiye le ju silẹ kọja awọn asẹnti, awọn ede-ede, tabi awọn agbegbe alariwo.
- Ohun afetigbọ sintetiki le jẹ aṣiṣe fun ọrọ ododo laisi isamisi to yege.
Next awọn igbesẹ
- Gba ifọkansi ti o fojuhan fun gbigba ohun, ti ẹda, ati ilotunlo.
- Didara idanwo kọja awọn agbohunsoke oniruuru ati awọn ipo abẹlẹ.
- Ṣetumo nigbati eniyan gbọdọ ṣe atunyẹwo tabi fọwọsi awọn abajade.
- Aami ohun sintetiki ki o tọju awọn igbasilẹ provenance fun iṣiro.
Tesiwaju Ṣiṣawari
Ìwé ìròyìn ọfẹ
Gba ìkànìyàn AI ojoojúmọ́
Ìtàn AI mẹta tó jẹ́rìí ní gbogbo owurọ́ ọjọ́ iṣẹ́, kọ́wé ní Gẹ̀ẹ́sì tó rọrùn. Ọfẹ́ fún gbogbo àkókò, kò sí ìpolówó.
Ìmeèlì kan ní gbogbo ọjọ́ iṣẹ́. Yọkúrò ní ìkànìyàn ní ìkànìyàn kan. A kò ta tàbí pín àdírẹ́sì rẹ.
Ṣe idanwo fun ara rẹ
Gba idanwo Source-Filter Vocoding ati WORLD
Ìdáhùn lẹ́sẹ̀kẹsẹ̀ lórí gbogbo ìdáhùn, àti iwe-ẹ̀rí tó lè pin pẹ̀lú ID tó lè jẹ́rìí lẹ́yìn tí o bá kọ́ ẹ̀kọ́.
Ṣe àtìlẹ́yìn ẹ̀kọ́ AI ọfẹ́. AI Understanding jẹ́ àjọ àìní èrè 501(c)(3) — kò sí ìpolówó, kò sí àkúnya owó, ní gbogbo àkókò. Ṣe ẹbun kan
Awọn ibeere ti a beere nigbagbogbo
Kini Source-Filter Vocoding ati WORLD?
Modeli source-filter pin si iṣiṣẹ ohun (speech excitation) kuro ni iṣeto oníṣọ (acoustic shaping) ti ọ̀nà ohun (vocal tract). WORLD jẹ́ vocoder ti o ṣe iṣiro ati dahun ohun (resynthesizes speech) nipa lilo iye igbohunsafẹfẹ ipilẹ (fundamental frequency), awọ-ara spectral (spectral envelope), ati aperiodicity. A le ṣe atunṣe awọn paramita wọnyi lọtọ ninu ilana iṣakoso ami (signal-processing workflow), botilẹjẹpe atunṣe ko jẹrisi pe aṣoju oluwo (speaker identity) tabi dandan (naturalness) yoo wa kanna.
Ninu awoṣe orisun-àlẹmọ ti ọrọ, kini 'àlẹmọ' naa ṣe aṣoju?
Àlẹmọ jẹ ariwo ti ngba ohun - irisi ọfun, ẹnu, ati imu ti o ṣe awọ ohun naa. Orisun ni ariwo-okun ariwo tabi ariwo afẹfẹ.
Awọn paramita mẹta wo ni agbaye vocoder yọ jade lati ọrọ?
AGBAYE sọ ọrọ sisọ sinu igbohunsafẹfẹ ipilẹ (F0), apoowe iwoye (timbre/filter), ati aperiodicity (ariwo dipo iwọntunwọnsi ohun orin).
Kini orukọ algoridimu AYÉ fun ṣiṣe iṣiro apoowe iwoye?
CheapTrick ṣe iṣiro apoowe iwoye didan ti o lagbara si awọn aṣiṣe kekere ni iṣiro ipolowo.
Kini iyipada paramita ti WORLD n fun laaye oluṣe atunṣe lati ṣe taara?
WORLD n ṣafihan awọn ọna F0 ati awọ-ara spectral (spectral-envelope) lọtọ, nitorinaa eto kan le tun ọkan wọn ṣe lakoko ti o n gbe ekeji duro; abajade ti a le ri ko jẹrisi pe o yoo tọju aṣoju oluwo ni kikun.
Igbọkanṣe wo ni o ṣe alaye WORLD daradara ju vocoder neural lọ?
WORLD n ṣe iṣiro awọn paramita ohun ti o han gbangba (explicit) ati ṣẹda ohun lati inu wọn, nigba ti awọn vocoder neural n kọ ẹkọ iṣẹda waveform; didara ati iṣiro iṣiro da lori awọn eto ati ọna lilo.
Tesiwaju kikọ
Jẹmọ awọn itọsọna
Awọn itọsọna diẹ sii ti a yan fun koko yii