eos_2016_spring team5 - voiceprint recognition system

15
EOS PROJECT REPORT: VOICEPRINT RECOGNITION SYSTEM 0450756 陳陳陳 0450758 陳陳陳

Upload: nctusee

Post on 14-Feb-2017

67 views

Category:

Engineering


3 download

TRANSCRIPT

Page 1: EOS_2016_Spring Team5 - Voiceprint recognition system

EOS PROJECT REPORT:VOICEPRINT RECOGNITION SYSTEM0450756 陳儷文 0450758 陳亭甫

Page 2: EOS_2016_Spring Team5 - Voiceprint recognition system

構思• 傳統門禁系統僅利用密碼或是感應卡,本專題可透過嵌入式作業系統達成密碼輸入

搭配聲紋辨識,以增加門禁之安全。

Embedded Operating

SystemEnter the Password

Page 3: EOS_2016_Spring Team5 - Voiceprint recognition system

HARDWARE ARCHITECTURE

Enter password

Page 4: EOS_2016_Spring Team5 - Voiceprint recognition system

PROGRAMMING FLOWCHART – TRAINING

MFCC

Speaker Training(GMM)

Speaker Model 1Speaker Model 2Speaker Model 3

Speaker Model n

.

.

.

.

.

.

.

Speaker Model Database

.

.

.

.

.

.

Hamming

Window

MFCCHammin

g Window

MFCCHammin

g Window

Eliminate noise

Endpoint

Pre-Emphasi

sPre-

Emphasis

Pre-Emphasi

s

Thread1:

Thread2:

Thread n:

Page 5: EOS_2016_Spring Team5 - Voiceprint recognition system

PROGRAMMING FLOWCHART – VERIFICATION

MFCC

Eliminate noise

Endpoint......

Hamming

Window

MFCCHammin

g Window

MFCCHammin

g Window

Verification

Verification

Verification

Pre-Emphasi

sPre-

Emphasis

Pre-Emphasi

s

Vote

Accuracy Rate

> 50%

Open

Close

Yes

No

Vote

Vote

Vote result

Thread1:

Thread2:

Thread n:

Page 6: EOS_2016_Spring Team5 - Voiceprint recognition system

READFILE (.WAV)

• Channel : 1

• Sampling Rate : 8000

• Bps : 8000

• Data Size : 39950Bytes (5secs)

• Data : WAV.data[data_size]

Page 7: EOS_2016_Spring Team5 - Voiceprint recognition system

ENDPOINT DETECTION

• unsigned char WAV.data[data_size] • 1 bytes ( 8bits )• About 120 ~ 170

• 只選擇 WAV.data[8000]~WAV.[n-7999]

• 同時將這些 data 中,連續 20 個資料皆介於 137 到 142 中間的 frame 去除,留下有明顯聲音資料的部分做分析。

Page 8: EOS_2016_Spring Team5 - Voiceprint recognition system

PRE-EMPHASIS

• Pre-Emphasis 的功能是將語音訊號 x(n) 通過一個高通濾波器y(n) = x(n) – ax(n-1)

Page 9: EOS_2016_Spring Team5 - Voiceprint recognition system

HAMMING WINDOWING

• 將每一個音框乘上漢明窗,以增加音框左端和右端的連續性。假設音框化的訊號為 S(n), n = 0,…N-1 。那麼乘上漢明窗後為 S'(n) = S(n)*W(n) ,此 W(n) 形式如下:

W(n, a) = (1 - a) - a cos(2*pi*n/(N-1)), 0≦n≦N-1

Page 10: EOS_2016_Spring Team5 - Voiceprint recognition system

MFCC

1. 將一訊號進行傅利葉轉換( Fourier transform )2. 將頻譜映射( mapping )至梅爾刻度,利用三角窗函數( triangular overlapping

window ) 3. 取對數( logarithm ) 4. 取離散餘弦轉換( discrete cosine transform ) 5. MFCC 是轉換後的頻譜係數

Page 11: EOS_2016_Spring Team5 - Voiceprint recognition system

GMM (高斯混合模型)• 根據獲得的 MFCC 值,將其當作一 13 維的 Vector ,利用 16 個高斯函數以及

100 個 training data ,建立出表示每個人不同的 Model

• 當 test data 進來時,便與已經建立好的各個模型去做比較,計算其機率值,判斷是否符合。

Page 12: EOS_2016_Spring Team5 - Voiceprint recognition system

SKILL

• Timer• Signaling• Process• Thread• System call and Execl• Audio I/O

Page 13: EOS_2016_Spring Team5 - Voiceprint recognition system

DIVISION OF THE WORK

• Structure• 陳亭甫、陳儷文

• Program• 陳亭甫 – Endpoint, Pre-emphasis, Hamming window, MFCC, ReadFile• 陳儷文 – GMM, Vote, Input (Audio control and Keyboard)

• Report and PPT• 陳亭甫、陳儷文

Page 14: EOS_2016_Spring Team5 - Voiceprint recognition system

DEMO TIME

Page 15: EOS_2016_Spring Team5 - Voiceprint recognition system

THANK YOU FOR YOUR ATTENTION