lifefeel / ims-speech

IMS-Speech

Installation

$ git clone --recurse-submodules [e-mail]:Pavel/ims-speech.git

download modelfile from https://kaldi-asr.org/models/m4

ln -s ../espnet/egs/korean/asr3/conf conf ln -s ../espnet/tools/kaldi/egs/aspire/s5/utils/run.pl run.pl ln -s ../espnet/tools/kaldi/egs/aspire/s5/utils utils ln -s ../espnet/tools/kaldi/egs/aspire/s5/steps steps ln -s ../espnet espnet

Copy Mode for VAD to dir. for VAD + ASR

cp ./models/en/conf/mfcc_hires.conf ./conf

Segmentation of about 40 sec from Large File

Shell Command

sh longSeg.sh or
python longWavSplit.py 001

• 입력: ./sample_data_16k/eighteen2
  • 001.wav
• 출력: ./sample_data_16k/eighteen2
  • 001Out.txt ; segmentation 정보
  • 001_eDir ; Dir. for Segmented files

VAD + ASR for each segmentation

Usage 1

Shell Command

sh example.sh or
python wav_split2.py  file002_e

• 출력: ./sampleOut
  • file001_eOut.txt ; segmentation 정보
  • file001_eDir ; Dir. for Segmented files

vadR1.sh 수행 : speech segmentation VAD

../espnet/egs/korean/asr3/decode.sh 수행: ASR

Changed from Original Version

• ./steps/segmentation/detect_speech_activityVad.sh
• ./steps/make_mfccVad.sh

Output

./sampleOut
   file002_eOut.txt

Usage 2 for fast speech segmentation Only(VAD)

$ ./vadR.sh file002_e.wav

Usage 3 for traditional speech segmentation Only (VAD)

$ ./vad.sh file002_e.wav

Output

./temp/segmentation/output_seg/segmentation

13 segmentation

50e9ae7561526f097f01d0c9c47a8de3-0000154-0000302 50e9ae7561526f097f01d0c9c47a8de3 1.54 3.02 50e9ae7561526f097f01d0c9c47a8de3-0000354-0000508 50e9ae7561526f097f01d0c9c47a8de3 3.54 5.08 50e9ae7561526f097f01d0c9c47a8de3-0000604-0000903 50e9ae7561526f097f01d0c9c47a8de3 6.04 9.04 50e9ae7561526f097f01d0c9c47a8de3-0001091-0001403 50e9ae7561526f097f01d0c9c47a8de3 10.91 14.03 50e9ae7561526f097f01d0c9c47a8de3-0001433-0001563 50e9ae7561526f097f01d0c9c47a8de3 14.33 15.63 50e9ae7561526f097f01d0c9c47a8de3-0001593-0001764 50e9ae7561526f097f01d0c9c47a8de3 15.93 17.64 50e9ae7561526f097f01d0c9c47a8de3-0001927-0002066 50e9ae7561526f097f01d0c9c47a8de3 19.27 20.66 50e9ae7561526f097f01d0c9c47a8de3-0002119-0002624 50e9ae7561526f097f01d0c9c47a8de3 21.19 26.24 50e9ae7561526f097f01d0c9c47a8de3-0002654-0002945 50e9ae7561526f097f01d0c9c47a8de3 26.54 29.45 50e9ae7561526f097f01d0c9c47a8de3-0002994-0003147 50e9ae7561526f097f01d0c9c47a8de3 29.94 31.47 50e9ae7561526f097f01d0c9c47a8de3-0003575-0003695 50e9ae7561526f097f01d0c9c47a8de3 35.75 36.95 50e9ae7561526f097f01d0c9c47a8de3-0003778-0003923 50e9ae7561526f097f01d0c9c47a8de3 37.78 39.23 50e9ae7561526f097f01d0c9c47a8de3-0003964-0004171 50e9ae7561526f097f01d0c9c47a8de3 39.64 41.71