AUTOMATIC SPEECH RECOGNITION OF LOW RESOURCE LANGUAGES

• Author(s) / Creator(s)

  • Allan Plesniarski

• This study focused on exploring ASR systems primarily for the transcription of the Totonac languages of Coatepec and Upper Necaxa. Best ASR transcription results were achieved using Meta Research MMS multilingual model with the Wave2vec ASR framework. The Totonac languages were transcribed with a reasonable Phoneme Error rate based on the Highland Totonac language trained into the Meta MMS model. The transcription accuracy of consonants is higher than vowels, giving a linguistic researcher an automatically transcribed template that can serve as the basis for manual fine-tuning of phonemes and word boundaries.

• Date created

  2024-04-01

• Subjects / Keywords

  • ASR
  • automatic speech recognition
  • low resource language
  • Totonac

• Type of Item

  Research Material

• DOI

  https://doi.org/10.7939/r3-dzx9-gf85

• License

  Attribution-NonCommercial 4.0 International