Bangla-align is a toolkit for force-aligning Bangla/Bengali. It is based on the Montreal Forced Aligner. INPUT: 1) audio files, 2) corresponding "sentence-level" transcription as Praat TextGrids OUTPUT: Praat TextGrids with Phoneme-level annotations.
Here are the major steps of force-algning Bangla speech:
-
Place the audio and their corresponding transcription TextGrids inside the folder "input_audio_and_transcription_tgs"
-
Start a terminal window and change directory to "bangla-align*"
-
type the following command in the terminal to run run Script_1...
$ bash Script_1_generate_dictionary.sh
-
A file named "temp_dict_bangla_ortho.dict" will be generated; inspect this file and correct any error in pronunciation of any word
-
If any word was not found in the master phoneme dictionary, another file named "oov_with_suggestions.txt" will also be generated; inspect this file and correct errors
-
type the following command in the terminal to run run Script_2...
$ bash Script_2_mfa_align_and_postprocessing.sh
-
The final output TextGrids will be generated under a new folder named "Z_final_aligned_tgs"