中原大學電機工程學系
Loading
專題成果首頁>專題成果
104學年度
語音辨認
指導老師:陳德鑫   組長:金聖格   組員:王俞強、洪偉國
進行語音辨識前,有許多步驟預先處理,像是過零率,端點界定,自相關函數,動態時間軸校正,抓取基本頻率。過零率是對氣音和濁音的最基本辨識,氣音(及雜訊)的過零率通常均大於濁音。端點界定是為了避免待處理的音訊起始點,終點,被雜訊混淆。端點界定在時域處理時,資料量較小。在頻域則較大。自相關函數可以透過聲音的比對,找出片段的基本頻率。而動態時間軸校正,是透過路線的最佳化,讓時間軸不致偏差過大。
經處理過後的語音,則可以抓取特徵,進行比對。在特徵擷取的步驟,將帶入小波轉換,梅爾倒頻譜,線性預測係數等函數。建立語音模型資料,以待進行比對驗證,最後整合出的辨識器,則可以拿來驗證待測語音。
To deal with speech recognition,there are lot of thing needed to be done first.Such like Zero Crossing Rate, End-point Detection, Autocorrelation, Dynamic Time Warping ,and catching the fundamental frequency.we can compute the ZCR of the sound to recognize Vowel and Consonant.The ZCR of Consonant usually bigger than Vowel.Though the EPD,the begin and the finish of a sound could be defined.it would be less data using the time-domain,and large data by freq-domain.
Using the Autocorrelation, we can find the partial fundamental frequency.And the DTW can let the bias down.To do the know the comparison of the sound.We would run functions like wavelet transform, MFCC, linear prediction.To correct the sound,we also need to build lots of model. Finally, the real sound could be recognized
我們決定將壓電陶瓷晶片和鞋子做結合。
 
Primobolan Depot For Sale Hygetropin 10iu Dosage Jintropin Kigtropin Vs Somatropin Turinabol For Sale Testosterone Propionate Cycle Anavar for sale Igtropin 100mcg Price Android Pills Stanazol Suspension Side Effect Buy HCG 5000 iu Vials Anadrol Oxymetholone Buy Ansomone Growth Hormone Anadrol For Sale Testosterone Enanthate 250mg Testosterone Cypionate For Sale Deca Durabolin Side Effects Stanozolol Suspension Injection 20ml Winstrol for Sale trenbolone acetate
 
 
32023 桃園市中壢區中北路200號 No-.200, Zhongbei Rd., Zhongli City, Taoyuan County 320, Taiwan(ROC) TEL:03-2654801 FAX:03-2654899