################################################################################################# This README.txt explains steps to obtain (a) wav/, (b) etc/txt.done.data, (c) lab_{wd/phn}_level/ for building TTS system, given just audio (BIG_FILE.wav) without corresponding text. The BIG_FILE.wav is split into shorter wavefiles using sil. detection algo, and are copied to wav/ directory. The etc/txt.done.data and lab/ directory are produced by decoding audiobook using Librispeech acoustic model in Kaldi framework. ################################################################################################# (1) Downloading Kaldi (2) Installing Kaldi on Linux (3) Set variables and execute run.sh
(1) Downloading Kaldi -- Make sure that subversion (svn) is installed -- Type the following svn co https://svn.code.sf.net/p/kaldi/code/trunk kaldi-trunk -- In case, above command throws error, edit ~/.subversion/servers to make [global] http-proxy-host = proxy.iiit.ac.in (line 144) http-proxy-port = 8080 (line 145)
(2) Installing Kaldi on Linux -- A folder named "kaldi-trunk" must have been created in the present directory -- cd kaldi-trunk/tools -- make -j <num_free_CPUs> // use -j option for faster installation -- cd ../src -- ./configure -- make depend -j <num_free_CPUs> -- make -j <num_free_CPUs> -- Kaldi installation is complete. -- Add .../kaldi-trunk/src/*bin, .../kaldi-trunk/tools/openfst-1.3.4/src/bin and .../kaldi-trunk/tools/irstlm/bin to $PATH in ~/.bashrc file
(3) Set variables and execute run.sh -- Unzip prep_data_for_TTS.zip -- Set KALDI-ROOT variable to .../kaldi-trunk (abs. path where Kaldi is installed) in prep_data_for_TTS/scripts/run_kaldi/path.sh file -- cd prep_data_for_TTS/, set the following variables in run.sh, and run BIG_FILE="./BIG_FILE.wav" expt_name="EMMA" spk_gender="f" // (m/f) num_free_CPUs=20 Ngram=1 adapt="yes" // (yes/no). If yes, transcript produced using // Librispeech models is used for adapting the model to // audiobook data.