3.2 Changelog

3.2.1

• Changed unicode normalization to default to composed forms unless overridden by --unicode_decomposition true

3.2.0

• Added --subset_word_count parameter to Train a new acoustic model (mfa train) to add a minimum word count for an utterance to be included in training subsets

• Added --minimum_utterance_length parameter to Train a new acoustic model (mfa train) to add a minimum word count for an utterance to be included in training at all

• Improved memory usage in compiling training graphs for initial subsets

• Add support for transcription via whisperx and speechbrain models

• Update text normalization to normalize to decomposed forms

• Compatibility with Kalpy 0.6.7