lmaxwell / force-alignmet

do force-alignment on mandarin corpus

Hi there 👋

• 🔭 I’m currently working on audio self-supervised learning
• I produce music with ableton live and programming languages like ChucK
• 📫 How to reach me: https://lmaxwell.github.io

force-alignmet's Issues

这项目能否对齐到声韵母级别？

thx in advance

LOpen: Unable to open label file mfc/1.lab

Hey抱歉我不熟悉HTK这个工具包，我使用merlin/tools/complie_htk.sh脚本安装了HTK，按照tutorial，将一个wav(44100khz)和相应的words.mlf放到相应的位置，运行时遇到卡在了如下错误。

  ERROR [+6510]  LOpen: Unable to open label file mfc/1.lab
 FATAL ERROR - Terminating program /home/jackie/proj/merlin/tools/bin/htk/HVite

此外，还有以下这个错误，不知道是否因为这个原因导致的

No HTK Configuration Parameters Set

部分运行日志

perl script/align.pl
make phone level transcrition
generate SCP file
Extract features
 wav/1.wav -> mfc/1.mfc 
make proto hmm 
/home/jackie/proj/merlin/tools/bin/htk/HCompV -A -D -T 1 -C config/train.cfg -M mmf -S general/train.scp -m -f 0.001 mmf/proto 

HTK Configuration Parameters[7]
  Module/Tool     Parameter                  Value
#                 NUMCEPS                       12
#                 NUMCHANS                      20
#                 PREEMCOEF               0.970000
#                 USEHAMMING                  TRUE
#                 WINDOWSIZE         200000.000000
#                 TARGETRATE          50000.000000
#                 TARGETKIND            MFCC_0_D_A

Calculating Fixed Variance
  HMM Prototype: mmf/proto
  Segment Label: None
  Num Streams  : 1
  UpdatingMeans: Yes
  Target Direct: mmf
1248 speech frames accumulated for stream 1
Updating HMM Means and Covariances
Output written to directory mmf
Var floor macros output to file mmf/vFloors