Ọrọ sisọ pẹlu RNNoise
RNnoise jẹ aami kekere, nẹtiwọọki nkankikan ti o yara ti o yọ ariwo lẹhin lati ọrọ ni akoko gidi.
Akopọ
Created by Xiph.Org's Jean-Marc Valin, it pairs classic signal processing with a small recurrent network so it runs on ordinary CPUs and even embedded devices.
Jin Dive
RNnoise, ti a tu silẹ ni ọdun 2017, jẹ apẹrẹ fun idinku ariwo-kekere ni awọn ipe ohun. Dipo ki o kọ ohun gbogbo ni ipari-si-opin, o pin ọrọ si bii awọn ẹgbẹ igbohunsafẹfẹ 22 ti a ṣe apẹrẹ lori eti eniyan (iwọn-iwọn bi epo igi) ati pe o nlo nẹtiwọọki aifọkanbalẹ loorekoore pẹlu Awọn Ẹka Loorekoore Gated lati ṣe iṣiro ere kan (0 si 1) fun ẹgbẹ kọọkan fun fireemu kan. Awọn anfani yẹn dinku awọn ẹgbẹ alariwo lakoko ti o tọju awọn ẹgbẹ ti o jẹ gaba lori ọrọ-ọrọ. Àlẹmọ ọ̀rọ̀ àfikún kan ń fọ ariwo tí ó ṣẹ́ kù láàárín àwọn ìṣọ̀kan ti ọ̀rọ̀ sísọ. Gbogbo awoṣe ni aijọju awọn iwọn 85,000, nṣiṣẹ yiyara ju akoko gidi lọ lori ipilẹ Sipiyu kan, ati pe o jẹ orisun ṣiṣi labẹ iwe-aṣẹ BSD, eyiti o jẹ idi ti o fi ṣepọ si awọn iṣẹ akanṣe bii ilolupo koodu codec Opus, Mumble, ati OBS Studio.
Imọ-imọ-ẹrọ
Yiyan apẹrẹ bọtini n ṣiṣẹ lori awọn anfani ẹgbẹ oye dipo awọn apoti iwoye aise. Nipa sisọ asọtẹlẹ ~ 22 awọn iye ere fun fireemu kan, nẹtiwọọki GRU duro kekere ati yago fun awọn ohun-ọṣọ ariwo-orin ti o wọpọ ni awọn ọna iyokuro ti ogbo. Awọn ẹya ara ẹrọ ti a fi ọwọ ṣe (awọn okun agbara ẹgbẹ, akoko ipolowo, ibaramu ipolowo) jẹun nẹtiwọọki, idapọ imọ DSP pẹlu ẹkọ. Iṣagbejade iṣẹ ohun lọtọ ṣe iranlọwọ awọn anfani ẹnu-ọna lakoko awọn fireemu ariwo-funfun.
Ipa Ilana
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.
Ọjọ iwaju ti Ọrọ sisọ pẹlu RNnoise
RNnoise ṣe atilẹyin igbi ti iṣẹ imudara akoko gidi iwuwo fẹẹrẹ; iwadii arọpo rẹ (PercepNet, DeepFilterNet) n gbe didara ga julọ lakoko ti o tọju awọn isuna Sipiyu kekere. Reti awọn apanirun lati fi sabe taara sinu awọn agbekọri, awọn iranlọwọ igbọran, ati awọn eerun apejọ, lati darapọ pẹlu ifagile iwoyi ati imukuro, ati lati lo oye ati paapaa awọn ibi-afẹde ipilẹṣẹ. Ohunelo DSP-plus-small-nẹtiwọọki arabara si wa ni ipa nibikibi ti aisi kekere, agbara kekere, ati ọrọ iwe-aṣẹ orisun-ìmọ diẹ sii ju iwọn awoṣe aise lọ.
Real-World imuse
Dinku clatter keyboard ati fan hum lakoko awọn ipe fidio ni awọn ohun elo ti o di RNnoise.
Lilọ soke gbohungbohun ṣiṣan ṣiṣan ni ile-iṣere OBS nipasẹ àlẹmọ-imukuro ariwo RNnoise ti a ṣe sinu.
Imudarasi oye ti iwiregbe ohun ni awọn ere ati awọn irinṣẹ VoIP bii Mumble lori ohun elo agbara kekere.
Ṣiṣeto awọn gbigbasilẹ aaye alariwo ṣaaju ki idanimọ ọrọ isale gba ifihan mimọ kan.
Awọn ewu & Awọn ọna iṣọ
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.
Ilana Ilana imuse
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
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Speech Denoising with RNNoise quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Itọsọna atẹle
Iyapa Ọrọ ati Amulumala Party Isoro
Awọn ibeere ti a beere nigbagbogbo
What is Speech Denoising with RNNoise?
RNnoise jẹ aami kekere, nẹtiwọọki nkankikan ti o yara ti o yọ ariwo lẹhin lati ọrọ ni akoko gidi. Ti a ṣẹda nipasẹ Xiph.Org's Jean-Marc Valin, o sopọ sisẹ ifihan agbara Ayebaye pẹlu nẹtiwọọki loorekoore kekere kan nitorinaa o ṣiṣẹ lori awọn CPUs lasan ati paapaa awọn ẹrọ ti a fi sii.
Kini RNnoise ni akọkọ asọtẹlẹ fun fireemu kukuru kọọkan ti ohun?
RNnoise ṣe iṣiro awọn anfani ẹgbẹ-ẹgbẹ kan ti o dinku awọn ẹgbẹ ti o jẹ gaba lori ariwo lakoko ti o tọju awọn ti o jẹ gaba lori ọrọ, dipo ki o ṣiṣẹ lori gbogbo awọn abala iwoye.
Iru nẹtiwọki nkankikan wo ni o wa ni mojuto ti RNnoise?
RNnoise nlo nẹtiwọọki ti nwaye loorekoore ti a ṣe pẹlu Gated Recurrent Units, fifun ni iranti igba diẹ lakoko ti o wa ni kekere.
Kini idi ti RNnoise n lo awọn ẹgbẹ igbohunsafẹfẹ oye dipo awọn apoti aise aise?
Sisọtẹlẹ ~ 22 awọn anfani ẹgbẹ nikan jẹ ki nẹtiwọọki kere, yiyara, ati ki o kere si awọn ohun-ọṣọ ariwo-orin ti o kọlu awọn ọna-bin.
Ni aijọju bawo ni awoṣe RNnoise ṣe tobi to?
RNnoise ni lori aṣẹ ti awọn iwọn 85,000, kekere to lati ṣiṣe yiyara ju akoko gidi lọ lori mojuto Sipiyu kan.
Kini ipa ti àlẹmọ ipolowo ni RNNoise?
Ajọ comb/pitch nilokulo ọna ibaramu ti ọrọ sisọ lati dinku ariwo ti o joko laarin awọn irẹpọ, ni ibamu awọn anfani ẹgbẹ naa.