Cach Thay Doi Cao Tien Giong Noi Cua Ban Theo Thoi Gian Thuc
Vocal pitch changer la mot trong nhung cong cu nghe co ve dang quang cho den khi ban thuc su co gang xay dung no — sau do ban nhan ra co bao nhieu xu ly tin hieu o giua “tang cao tien” va “van nghe nhu con nguoi”. Cho du ban muon giong noi radio sau hon de streaming, am cao hon cho nhan vat, hoac chi muon hieu phan mem streaming cua ban dang lam gi o dang sau, huong dan nay bao phu toan bo: ly thuyet DSP, cac cai dat thuc su quan trong, va thiet lap tung buoc cua thuc te trong VoxBooster cho Discord, tro choi, va OBS.
TL;DR
- Pitch shifting thay doi tan so khong thay doi toc do — su phan biet nay quan trong cho do tre va chat luong.
- Cac thuat toan phase-vocoder va time-domain moi cai co cac trade-off; biet cai nao cong cu cua ban su dung giai thich cac do tre ma ban nghe.
- Semitone la don vi dung; ±3–6 semitone bao phu hau het cac bien doi giong noi thuc te.
- Formant correction khong tuy chon neu ban muon nghe nhu con nguoi.
- VoxBooster dang ki mic ao tieu chuan (WASAPI, khong driver kernel) ma bat ki ung dung nao co the chon.
- Do tre duoi 10 ms co the dat duoc tren hardware hien dai voi cac cai dat buffer dung.
Pitch Shifting Thuc Su Lam Gi
Khi ban tang toc do phat mot ban ghi tren bang, cao tien tang. Lam cham, cao tien giam. Moi quan he giua toc do va cao tien ay la phuong phap naive — va no vo dung cho cong viec giong noi theo thoi gian thuc vi no cung keo dai hoac nen thoi gian, khien loi noi tro nen khong the hieu duoc.
Pitch shifting thuc te tach cao tien khoi thoi gian. Tin hieu duoc chia thanh cac doan ngan chong cheo, moi doan duoc dich chuyen tan so (hoac thong qua tac dong pho trong mien tan so hoac thong qua thu tro toc do phat trong mien thoi gian), va cac doan duoc khoan lai o nhip do ban dau. Nguoi nghe nghe giong noi da thay doi cao tien chinh xac voi toc do ban noi.
Su tach nay la toan bo thach thuc ky thuat. No cung la ly do tai sao pitch shifting chat luong cao co chi phi CPU khong tam thuong va tai sao cac trien khai re tien tao ra cac do tre “kim loai” hoac “robot” dac trung.
Phase Vocoder: Thuat Toan Thong Tri
Phase vocoder la gi va tai sao no quan trong cho am thanh theo thoi gian thuc?
Phase vocoder chuyen tin hieu am thanh vao mien tan so su dung Short-Time Fourier Transform (STFT), dich chuyen tung bin tan so bang mot he so khong doi (vi du, ×1.189 cho +3 semitone, vi 2^(3/12) ≈ 1.189), sau do tai tao tin hieu mien thoi gian voi STFT dao. Vi tan so va pha duoc theo doi rieng biet, thoi gian co the duoc giu nguyen. “Phase” trong ten dung toi theo doi deu tam pha duoc yeu cau de tranh deo nhoa cac chuyen di tren cua so overlap-add tong hop.
Cac tham so chinh:
- Kich thuoc cua so FFT — Cua so lon hon cho do phan giai tan so tot hon (cao tien sach hon) nhung co do tre hon. Cua so 2048-point o 48 kHz them khoang 42 ms do tre tu chinh cua so; cua so 512-point cat xuong ~10 ms nhung gioi thieu nhieu mo do tren pho tan so hon.
- Hop size — Khoang cach cua so phan tich tien tien moi khung. Hop nho hon = chong cheo hon = tru truot hon nhung CPU nang hon.
- Phase locking — Mot so trien khai khoa pha cua cac dinh cao tan so lai voi nhau, giam “phasiness” tren cac nguyen am dai ho hon voi chi phi CPU hoi it hon.
Cho viec su dung theo thoi gian thuc, trade-off la thang thang: cua so nho hon cho do tre thap hon, cua so lon hon cho chat luong. Cac cong cu tot phuc ta la mot cai ve chat luong/do tre dung thay vi tham so FFT that.
Co rong rai tai lieu hoc thuat ve thiet ke phase-vocoder — tap chi co ban boi Flanagan va Golden (1966) va cac cong viec sau nay boi Laroche va Dolson la diem khoi dau tot neu ban muon sang sao. Bai viet phase vocoder Wikipedia la tong quan hop ly ve toan hoc.
Time-Domain Pitch Shifting: PSOLA va Cac Bien The
Mot ho thuat toan khac hoat dong trong mien thoi gian thay vi mien tan so. Pho bien nhat la PSOLA (Pitch-Synchronous Overlap-Add), which:
- Phat hien chu ky co ban (chu ky cao tien) cua tin hieu phat thanh.
- Trich xuat hat co kich thuoc chu ky cao tien.
- Thap lai tren khoang cach khac de thay doi cao tien.
PSOLA rat hieu qua CPU va tao ra cac ket qua nghe rat tu nhien tren loi noi don am sach — chinh xac nhung gi ma voice changer lam. No gan nhu khong the tien hanh voi phu am khong phat thanh (fricative nhu /s/, /f/) va dau vao om, trong do chu ky cao tien khong xac dinh. Nhieu voice changer thuong mai su dung lai: PSOLA cho loi noi phat thanh, FFT cho tat ca.
Khoa hoc thuc te: neu ban nghe cac do tre dac biet tren cac am sibilant (s, sh, f, th) nhung cac nguyen am nghe sach, ban co the dang su dung cong cu dua tren PSOLA. Neu cac do tre nam duong — tia kim loai tren tat ca am thanh — co the la thuat toan FFT don gian hon ma khong co phase locking dung.
Semitone: Don Vi Dung Cho Pitch Shifting
Tan so duoc do bang Hz, nhung khoang cach nhan thuc giua cac cao tien la logarithmic. Semitone la 1/12 cua mot octave, tuong ung voi ti le tan so 2^(1/12) ≈ 1.0595. Dieu do co nghia:
| Dich Chuyen Semitone | Cong Thuc Nhan Tan So | Hieu Ung Nhan Thuc |
|---|---|---|
| +1 | ×1.06 | Gan nhu khong thay |
| +3 | ×1.19 | Cao tien mot it, van tu nhien |
| +6 | ×1.41 | Cao tien nhan thay, bien gioi chip dinh ma khong sua loi formant |
| +12 | ×2.00 | Octave day du len — ro rang duoc xu ly |
| -3 | ×0.84 | Sau mot it, tap tin |
| -5 | ×0.75 | Sau nhan thay, tot cho giong noi radio |
| -8 | ×0.63 | Rat sau, robot ma khong sua loi formant |
| -12 | ×0.50 | Octave day du xuong — ro rang tong hop |
Cac bien doi giong noi thuc te thuc hang o khoang ±2 den ±7 semitone. Vuot qua dieu do, bu cong formant tro thanh phu hop de giu ket qua nghe nhu giong noi con nguoi thay vi hieu ung robot.
Luu y rang nhieu cong cu hien thi cao tien trong semitone, cent (1/100 semitone), hoac thim thi la ti le tan so thot. VoxBooster su dung semitone lam don vi chu yeu, la truc quan nhat cho cong viec giong noi.
Formant: Tai Sao Cao Tien Khong Trong Mot Minh
Khi ban thay doi cao tien ma khong co thuoc formant, ban nhan duoc hieu ung chipmunk hoac ogre dien hinh. Day la ly do.
Giong noi con nguoi co hai thanh phan chinh: nguon (am om cua day thanh am, xac dinh cao tien) va bo loc (cac khoang resonan cua hong va mieng, hinh thanh mau sac pho va xac dinh ky thuat cua giong noi). Cac dinh resonan cua bo loc goi la formant.
Khi cao tien tang 6 semitone, nguon di len. Nhung vocal tract khong thay doi do dai vat ly — nen formant o y nguyen. Ket qua nghe sai vi nao khong su dung ti le giua tan so co ban va formant de danh gia kich thuoc nguoi noi. Cao tien cao voi formant thap nghe nhu con vat nho trong than the lon (chipmunk voi hong to).
Formant correction di chuyen cac dinh formant ti le voi dich chuyen cao tien, bo sung dieu se xay ra neu mot nguoi co day thanh am tu nhien cao hon (nguoi noi nho hon) noi dieu tuong tu. Ket qua nghe nhu mot nguoi thuc su khac thay vi phien ban duoc xu ly cua ban.
Trong VoxBooster, sua loi formant duoc kich hoat theo mac dinh khi ban chon tien set, va ban cung co the canh chinh no bang cach su dung knob Formant rieng biet cung voi knob Pitch. Ca hai co the di chuyen doc lap — huu ich khi ban muon the cua mot giong noi sau hon nhung cao tien cao hon mot it, hoac nguoc lai.
Sau Hon vs. Cao Tien Hon: Cac Canh Chinh Thuc Te
Di Sau Hon (Nam Tinh, Radio, Quai Vat)
Cho mot giong noi sau hon van tu nhien:
- Pitch: -3 den -5 semitone
- Formant: -1 den -2 semitone (di chuyen formant it hon cao tien cho ket qua tu nhien)
- Noise suppression: On — cac giong noi sau hon len giong hoi that hon
- Compression: Nhe (3:1 ratio) de can bang dong luc
Sai lam pho bien la di sau toc quang do. -5 semitone da la su bien doi co y nghia. O -7 hoac duoi do, ban gan nhu luon luon can bo sung formant cua it nhat -2 semitone hoac ket qua nghe nhu hang tom thay vi sau.
Cho hieu ung quai vat hoac robot day du, ban muon do tre duoc phat da — nen tat formant linking va day cao tien xuong -8 hoac -10. Xem huong dan hieu ung robot voice va bai viet hieu ung radio voice cho cac preset chuyen dung.
Di Cao Tien Hon (Nu Tinh, Chipmunk, Nhan Vat)
Cho mot giong noi cao tien va nhe hon:
- Pitch: +3 den +6 semitone
- Formant: +2 den +4 semitone (khop hoac vuot qua dich chuyen cao tien cho giong noi nu/tre em thuyết phuc)
- Sibilance: Canh do tren cac am /s/ duoc phat da — de-esser hoac cat tan so cao nhe o tren 8 kHz giup
- Breath noise: Ro rang hon o cao tien cao hon; su dung noise gate
Cho hieu ung chipmunk co chu y, dich chuyen cao tien +8 den +12 voi formant bi khoa hoac dich chuyen nieu hon — chinh xac tinh huong formant khong khop duoc mo ta o tren, su dung co chu y. Xem hieu ung chipmunk voice cho tung buoc.
Do Tre: Dieu Nao Gay Ra Va Cach Giam Thieu
Pitch shifting theo thoi gian thuc them do tre tu hai nguon: do tre thuat toan (cua so phan tich) va do tre driver/buffer.
Do tre thuat toan khong the giam duoc cho thuat toan va kich thuoc cua so nhat dinh. FFT 512-point o tan so mau 48 kHz tao ra ~10.7 ms cua so. Them hop 256 mau, va ban dang nhin 5-11 ms do tre thuat toan khong the tranh khoi, tuy thuo vao trien khai. Mot so thuat toan mien thoi gian co the chay o do tre thap hon vi ho xu ly hat ngan hon.
Do tre buffer tuy thuo vao hardware. O buffer 128-mau (48 kHz), ban them 2.7 ms tren moi buffer trong chuoi. Cac chuoi dien hinh lien quan den hai buffer (dau vao va dau ra), nen ~5 ms. Buffer lon hon (1024+ mau) on dinh hon nhung them ~21 ms moi cai.
Do tre tong the co the dat duoc trong thiet lap duoc cau hinh tot: 8–15 ms. VoxBooster duoc thiet ke de o duoi 10 ms do tre them tren hardware co the xu ly buffer WASAPI 128-mau.
Cac tip thuc te de giam do tre toi thieu:
- Dat thiet bi am thanh Windows thanh 48 kHz, 24-bit — khop voi tan so xu ly noi bo cua VoxBooster
- Su dung che do WASAPI doc quyen neu thiet lap cua ban cho phep
- Dong cac phan mem am thanh khac (DAW, ung dung giong noi khac) co the cam giu thiet bi am thanh
- Tat cac audio enhancement cua Windows tren thiet bi microphone cua ban (nhan phai > Properties > Enhancements > Disable all)
- Su dung tai nghe day co theo ca theo thay vi Bluetooth — am thanh BT them 40–200 ms doc lap tu phan mem
Tung Buoc: Thiet Lap Pitch Shifting trong VoxBooster
1. Cai Dat va Mo VoxBooster
Tai tu voxbooster.com/download va chay trinh cai dat. VoxBooster dang ki microphone ao (thiet bi WASAPI tieu chuan, khong driver kernel). Thi nghiem mien phi 3 ngay cap quyen truy cap day du vao tat ca cac hieu ung bao gom pitch shifting va dieu khien formant.
2. Chon Thiet Bi Dau Vao Cua Ban
Mo VoxBooster va trong cua so chinh, chon microphone vat ly cua ban lam thiet bi dau vao. Neu ban co mic USB, chon theo ten. Neu ban co audio interface, chon dau vao WASAPI cho thiet bi do.
3. Dieu Chinh Pitch Shift
Nhan tab Voice Effects. Ban se thay knob Pitch (semitone) va knob Formant. Dat cao tien vao gia tri muc tieu cua ban — bat dau voi -4 cho giong noi sau hon hoac +4 cho giong noi cao hon. Dieu chinh formant theo huong tuong tu nhung it hung hon (vi du, -2 den -3 formant cho -4 pitch).
Dong ho thoi gian thuc cho thay muc am thanh duoc xu ly cua ban. Noi va xem no phan ung.
4. Dat VoxBooster Lam Dau Vao Trong Ung Dung Cua Ban
Discord: Settings > Voice & Video > Input Device > chon VoxBooster Virtual Mic. Xem huong dan thiet lap voice changer Discord day du cho anh chup man hinh.
OBS: Sources > Audio Input Capture > them VoxBooster Virtual Mic. Ngoai ra, su dung mixer am thanh OBS de dinh tuyen thiet bi VoxBooster lam nguon giam sat. Tai lieu OBS ve thiet lap am thanh bao phu cac tuy chon dinh tuyen.
Game: Hau het cac tro choi su dung thiet bi giao tiep mac dinh cua Windows. Dat VoxBooster Virtual Mic lam thiet bi giao tiep mac dinh trong cai dat Suara Windows (nhan phai bieu tuong loa > Sound settings > Input).
5. Kiem Tra va Tuy Chinh
Su dung bot Echo Test Discord hoac giam sat OBS de nghe ban than. Cac van de pho bien va sua chua:
- Am thanh robot/kim loai: Giam luong dich chuyen cao tien, hoac kich hoat sua loi formant neu tat
- Chipmunk o cao tien cao: Tang dich chuyen formant de khop hoac vuot qua dich chuyen cao tien
- Dau ra om: Kich hoat noise suppression trong chuoi hieu ung VoxBooster
- Clipping: Ha am luong microphone cua ban trong Windows; gioi han VoxBooster se bat cac dinh nhung ban muon dau vao sach
6. Luu Preset
Sau khi ban co cac cai dat ma ban thich, luu preset trong VoxBooster de ban co the chuyen doi giua giong noi binh thuong va phien ban pitch-shifted voi mot nhan (hoac hotkey). Lien ket hotkey dac biet huu ich cho luong livestream.
Pitch Shifting vs. Cac Hieu Ung Giong Noi Khac
Pitch shifting thuong duoc ket hop voi cac hieu ung khac cho cac hieu ung nhan vat toan dien hon. Day la cach cac hieu ung chinh tuong tac:
| Hieu Ung | Dieu No Lam | Ket Hop Tot Voi Cao Tien? |
|---|---|---|
| Pitch shift | Thay doi tan so co ban | — (tam cua hau het cac giong noi nhan vat) |
| Formant shift | Thay doi ky thuat vocal tract | Luon duoi dao voi cao tien |
| Reverb | Them khong gian/khong gian | Tot cho cac giong noi radio/announcer |
| Distortion | Them bao hoa hoa hoc | Cac giong noi demon/robot |
| Noise gate | Cat im lang/breath noise | Luon huu ich |
| EQ | Tang/giam cac band tan so | Tuy chinh thang sau cao tien |
| Compression | Can bang dong luc | Streaming/broadcasting |
| Noise suppression | Loai bo om xung quanh | Luon huu ich |
De kham pha cac preset hieu ung cu the, trang tinh nang voice effects co danh sach day du nhung gi ma VoxBooster bao gom.
So Sanh Cac Cong Cu Vocal Pitch Changer
Neu ban dang danh gia cac tuy chon, day la so sanh thang thang cua cac cong cu chinh trong khong gian nay:
| Cong Cu | Thoi Gian Thuc? | Dieu Khien Formant? | Mic Ao? | Do Tre | Gia |
|---|---|---|---|---|---|
| VoxBooster | Co | Co (doc lap) | Co (WASAPI) | <10 ms | Thi nghiem + tra phi |
| Voicemod | Co | Han che | Co | ~15–25 ms | Freemium |
| MorphVOX | Co | Co Ban | Co | ~20 ms | Thi nghiem + tra phi |
| Clownfish | Co | Khong | Co | Bien Doi | Mien Phi |
| DAW + plugin | Co | Tuy Thuo Plugin | Qua loopback | 5–40 ms | Khac Nhau |
DAW (nhu Reaper hoac REAPER Lite) voi plugin cao tien chat luong cao cho phep linh hoat toi da nhung can thiet ke co y — dinh tuyen thong qua cap ao, quan ly cau hinh phien, chay DAW day du o background. Doi voi cac livestreamer va game thu muon thiet lap nhanh va hotkey tin cay, phan mem voice-changer chuyen dung la tuy chon thuc te hon.
Cac Van De Pho Bien va Sua Chua
Pitch shift nghe tot khong co lien hoan nhung ban be tren Discord nghe doi tre. Discord ap dung noise suppression cua no (dua tren Krisp). Dieu nay co the tuong tac voi am thanh pitch-shifted va them cac doi tre cua no. Tat xu ly kenh nhiem cua Discord (Settings > Voice > Advanced > Noise Suppression > None) va su dung noise suppression noi tro cua VoxBooster.
Cao tien thay doi nhung giong noi nghe rong hoac phasey. Phase vocoder smearing — co gang giam luong dich chuyen cao tien mot chut hoac chuyen sang che do chat luong khac. Cua so FFT lon hon (che do do tre cao hon) thuong giai quyet van de nay tren cac nguyen am keo dai.
Giong noi cua toi sau hon nhung mo co the noi la toi. Pitch shift mot minh khong thay doi cac mau luoi, nhip, hoac trai thuat. Cho ket qua it dang nhan duoc hon, ket hop pitch shift voi sua loi formant va reverb nhe. Mot so nguoi dung cung dieu chinh nhip noi co y thuc.
Co echo hoac feedback. Giam sat cua ban co the duoc kich hoat tren dau ra ao. Tat nghe thiet bi nay tren mic ao VoxBooster trong tinh chat am thanh Windows, va su dung giam sat noi bo cua VoxBooster (bieu tuong headphone).
Cau Hoi Thuong Dat
Vocal pitch changer la gi?
Vocal pitch changer la phan mem thay doi tan so co ban cua giong noi cua ban tang hoac giam theo thoi gian thuc, khong thay doi toc do phat. No hoat dong bang cach phan tich am thanh cua ban, transpose tung thanh phan tan so, va xuat ket qua voi do tre hap nhat — thuong duoi 10 ms tren cac cong cu chat luong cao.
Toi can bao nhieu semitone de nghe giong noi cua mot nguoi khac?
Dich chuyen 3 den 5 semitone xuong duoi tao ra giong noi sau hon; 4 den 6 semitone len tren cho am cao hon va nhe hon. Cac dich chuyen lon hon 8 semitone co xu huong nghe nhu robot tru khi ban cung bo sung formant. Cac ket qua thuyết phuc nhat con nam trong khoang 2 den 6 semitone.
Pitch shifting co hoat dong ma khong can microphone ao khong?
Phan mem co the xu ly am thanh noi bo, nhung de su dung no trong Discord, tro choi, hoac cac ung dung streaming ban can mot thiet bi am thanh ao. VoxBooster cai dat microphone ao WASAPI tieu chuan ma bat ki ung dung nao cung thay nhu mot dau vao binh thuong — khong can driver kernel.
Pitch shifting theo thoi gian thuc se khien toi bi cam ban trong tro choi?
VoxBooster su dung WASAPI va dang ki nhu mot microphone ao binh thuong, nen cac he thong anti-cheat khong thay gi bat thuong. Khong co driver cap kernel duoc cai dat. Rui ro ve co ban bang khong, tuy nhien chinh sach tro choi rieng le ve sua doi am thanh co the khac nhau.
Formant correction la gi va toi co can no khong?
Formant correction dieu chinh cac du doan vocal tract (mau sac nhan thuc cua giong noi) doc lap voi cao tien. Ma khong no, dich chuyen cao tien len tren khien ban nghe nhu chipmunk; dich chuyen xuong duoi khien ban nghe khong tu nhien va day. Kich hoat formant linking mang lai ket qua tu nhien va con nguoi hon.
Lam the nao toi co the giam do tre khi pitch shifting theo thoi gian thuc?
Do tre tren duoc tu do tre thuat toan (kich thuoc cua so lon hon = khong do tre nhat nhung cham hon), kich thuoc buffer, va chi phi driver. Su dung audio interface chuyen dung hoac driver WASAPI cua mainboard, giu buffer VoxBooster o 128 hoac 256 mau, va dong cac phan mem am thanh nang khac.
Toi co the pitch-shift giong noi tren Discord ma khong co ung dung rieng khong?
Discord tu no khong co tinh nang pitch-shifting. Ban can phan mem chuyen dung nhu VoxBooster, dinh tuyen am thanh da xu ly thong qua mic ao ma Discord chon lam dau vao cua no. Thiet lap mat khoang hai phut.
Ket Luan
Real-time voice pitch changing la mot van de duoc giai quyet tu goc do ky thuat — cac thuat toan chin chuan va duoc hieu ro. Dieu phan biet cac cong cu tot tu trung binh la chat luong trien khai: pha deu, xu ly formant, quan ly do tre, va muc do dieu tuyen audio ao hoat dong voi cac ung dung ban thuc su su dung.
Hieu bieu do — semitone lam don vi dung, formant lam bo sung cua cao tien, kich thuoc cua so lam trade-off do tre/chat luong — cap cho ban tu vung de tuy chinh thiet lap cua ban thong minh thay vi chi quay cac knob cho den khi dieu gi nghe co the chap nhan.
VoxBooster ket hop mot dong co pitch phase-vocoder voi dieu khien formant doc lap, microphone ao WASAPI, va do tre sub-10 ms trong goi hang can khoang hai phut de thiet lap. Thi nghiem mien phi 3 ngay bao phu moi tinh nang, nen ban co the kiem tra tat ca cac cai dat cao tien va preset trc khi quyet dinh.
Tai VoxBooster — thi nghiem mien phi 3 ngay, Windows 10/11.