使用 HTK 的 HCopy 檔案就可以完成提取 mfcc 特征的工作。
1、需要準備内容如下: 一個配置檔案: XXX.conf 一個輸入輸出檔案:标明語音檔案的位址 和 對應輸出 mfcc 檔案的位址 注:以上兩個檔案要放在相同目錄下
2、給出配置檔案:自命名為hcopy.conf # # Example of anacoustical analysis configuration file # SOURCEFORMAT =WAV # Gives the format of the speechfiles TARGETKIND =MFCC_0_D_A # Identifier of the coefficients to use
# Unit = 0.1micro-second : WINDOWSIZE =250000.0 # = 25 ms = length of a timeframe TARGETRATE =100000.0 # = 10 ms = frameperiodicity
NUMCEPS = 12 # Number of MFCC coeffs (here from c1 toc12) USEHAMMING = T # Use of Hamming function forwindowing frames PREEMCOEF =0.97 # Pre-emphasiscoefficient NUMCHANS = 26 # Number of filterbankchannels CEPLIFTER = 22 # Lengthof cepstral liftering ENORMALIZE =T
NATURALWRITEORDER = T # TheEnd
3、給出輸入輸出檔案:自命名為hcopy.scp 截取部分如下: angry\201.wavmfcc_angry\201.mfc angry\202.wavmfcc_angry\202.mfc angry\203.wavmfcc_angry\203.mfc angry\204.wavmfcc_angry\204.mfc angry\205.wavmfcc_angry\205.mfc angry\206.wavmfcc_angry\206.mfc angry\207.wavmfcc_angry\207.mfc angry\208.wavmfcc_angry\208.mfc angry\209.wavmfcc_angry\209.mfc angry\210.wavmfcc_angry\210.mfc angry\211.wavmfcc_angry\211.mfc
4、在DOS視窗利用HCopy 檔案進行 mfcc特征提取 指令:HCopy -A -D-C hcopy.conf -S hcopy.scp 截圖如下:
這樣mfcc特征就提取成功了~