On the development of speech resources for the Mixtec language

Area of Science:

• Computational Linguistics
• Indigenous Language Revitalization
• Speech Technology

Background:

• Mixtec language faces endangerment due to urbanization and discrimination.
• Existing resources like dictionaries lack reliable native speaker pronunciation.
• Formal acoustic resources and speech technologies are absent for Mixtec.

Purpose of the Study:

• To develop crucial resources for the Mixtec language, including a speech database and an ASR system.
• To create a functional speech application for Mixtec language learners and speakers.
• To address the lack of technological tools for preserving and promoting the Mixtec language.

Main Methods:

• Created a speech database of traditional Mixtec narratives, phonetically and orthographically labeled.
• Developed a native speaker-adaptive automatic speech recognition (ASR) system using the database.
• Integrated the ASR system with a Mixtec-Spanish/Spanish-Mixtec text translator.

Main Results:

• A functional speech database and an integrated ASR and translation system were successfully developed.
• The speech application achieved a mean recognition/translation performance of up to 94.36% with non-native speakers.
• The developed resources proved effective for a multiuser speech application, despite the database's small size.

Conclusions:

• The developed speech database and ASR system are effective tools for Mixtec language preservation and accessibility.
• The project demonstrates the viability of creating speech technologies for endangered indigenous languages.
• This work provides a foundation for future research and development in Mixtec language technology.