rgriscom/Hadza-Aligner

Speech recognition model for the Hadza language # Hadza-Aligner This is a speech recognition model for the Hadza language which can be used to align phone-level transcriptions, created using the Prosodylab-Aligner forced alignment software (which itself is based on HTK). The model is still in development and currently only includes data from three speakers, sourced from the Hadza ELAR collection. # Data Organization The original labels can be found in the Labels folder (click here for accompanying audio recordings), and the aligned output labels can be found in the TextGrid_Output folder (for use with Praat acoustic analysis software). The individual .TextGrid files can be used with the input audio recordings, or the .Collection files can be used with chained audio files. The .yaml file is the model itself, which can be used with Prosodylab-Aligner to align additional transcriptions, and the .dict file contains a list of all words in the training data and their corresponding phones.