Robust yet lenient forced-aligner built on Kaldi. A tool for aligning speech with text.
Modified by Roman Scott for the Nofanity application.
Mac and Linux
Download the source code, install the dependencies (place an ffmpeg binary inside ext), and run ./install.sh. Then, inside the ext folder, rename k3 and m3 to gentleK3 and gentleM3, and run pyinstaller gentle.spec. Then, in the dist folder generated, run ./gentle to start the server.
Windows
Run git submodule init and git submodule update. Download an ffmpeg binary and place it inside the ext folder. Run the install_models.sh script as well.
Go inside the Kaldi folder and make the following changes in the code as shown in the commits: here, here, here, and here.
Then go inside kaldi/src and create a folder called gentle, and copy-paste the Makefile, k3-win.cc, and m3.cc from the ext folder into the new folder.
Go back to kaldi/src and to the parent Makefile: add the word gentle at the end of the line SUBDIRS = and at the end of the line after "The tools depend on all the libraries".
After this, follow the compilation instructions for Kaldi on Windows. NOTE: Set Runtime Options to Multithreaded (/MT) instead of Multithreaded DLL (/MD)
Grab your k3.exe and m3.exe files from the built solution, rename them to gentleK3.exe and gentleM3.exe, and move them into ext. Then run pyinstaller gentle.spec and, in the dist folder generated, run gentle.exe to start the server.
By default, the aligner listens at http://localhost:8765. That page has a graphical interface for transcribing audio, viewing results, and downloading data.
There is also a REST API so you can use Gentle in your programs. Here's an example of how to use the API with CURL:
curl -F "audio=@audio.mp3" -F "transcript=@words.txt" "http://localhost:8765/transcriptions?async=false"