Muna Speech to Text: Preserving a Language Through AI

Where is Muna Spoken?

Muna (also called Wuna) is spoken by around 300,000 people on Muna Island and parts of Buton in Southeast Sulawesi, Indonesia. It is an Austronesian language with a rich oral tradition, including epic poetry known as kabhanti. Despite its vitality, Muna has limited digital presence, making speech-to-text tools essential for documentation and media production.

Why Accurate Muna Transcription Matters

For Muna speakers, having reliable Muna speech to text technology means being able to convert spoken stories, sermons, and conversations into written form. This supports literacy in the native language, helps preserve cultural heritage, and enables subtitling of local content. Without AI transcription, manual transcription of Muna audio is slow and costly due to the scarcity of trained transcribers.

Challenges in Automatic Transcription of Muna

Vowel Length Distinction

Muna has five vowel phonemes, each with a long counterpart: /a aː e eː i iː o oː u uː/. Minimal pairs like watu (stone) and waatu (time) require the ASR to time the vowel duration precisely. Many general-purpose models fail here, but Speechyou’s acoustic model includes dedicated duration features.

Limited Training Data

As a low-resource language, Muna lacks large annotated speech corpora. Speechyou addresses this through transfer learning from closely related languages (e.g., Wolio, Cia-Cia) and active data collection from Muna-speaking communities. Continuous model updates ensure accuracy improves over time.

Dialectal Variation

The three main dialects — Gheo, Katobu, and Lasalimu — differ in phonology and vocabulary. For example, the Gheo dialect often replaces /s/ with /h/ in word-initial positions (e.g., sora vs hora for 'book'). Our platform allows dialect selection and automatic adaptation to local speech patterns.

Use Cases for Muna Transcription

• Oral History Preservation: Record elders reciting kabhanti and automatically transcribe them for archival. This helps linguists and community members alike.
• Subtitle Generation: Create Muna subtitles for local films, government announcements, and religious content. Speechyou exports SRT files ready for video platforms.
• Language Learning: Transcribe classroom lessons and create reading materials for students learning Muna at school.
• Research: Enable quick analysis of interview data for sociolinguistic or anthropological studies.
• Accessibility: Provide text alternatives for Muna radio broadcasts and live events for deaf community members.
• Media Production: Facilitate video editing by generating accurate transcripts that can be translated into Indonesian or English.

How Speechyou Addresses These Needs

Speechyou is built to handle the specific demands of Muna voice to text conversion. Our model supports:

• Real-time transcription for live speech
• Batch processing of long audio files (up to 10 hours)
• Custom vocabulary for dialectal words (e.g., kabhanti, sara)
• Export to multiple subtitle formats (SRT, VTT, TXT)

By combining transfer learning with community input, we achieve over 85% word accuracy on clean Muna speech — a significant improvement over generic ASR tools.

Conclusion

Muna is a living language that deserves modern tools to keep it vibrant in the digital age. Speechyou’s dedicated Muna automatic speech recognition fills a critical gap, enabling transcription, subtitling, and language preservation at scale. Whether you are a linguist, a content creator, or a community leader, Speechyou provides the accuracy and flexibility needed to work with Muna audio effectively.