Preserving the Voice of Miyako: AI-Powered Speech-to-Text for an Endangered Language

The Miyako Language: A Voice on the Edge of Silence

Miyako (Myaakufutsu) is a Ryukyuan language spoken on the Miyako Islands, a tropical archipelago southwest of Okinawa, Japan. With an estimated 10,000 speakers, it is classified as endangered by UNESCO. The language is distinct from Japanese, belonging to the Japonic family but with its own grammar, vocabulary, and sound system. For centuries, Miyako was transmitted orally through folk tales, songs, and daily conversation. Today, revitalization efforts depend on capturing these spoken words in written form, so they can be studied, taught, and shared.

Why Accurate Speech-to-Text Matters for Miyako

Transcribing Miyako by hand is slow and labor-intensive. Most native speakers are elderly, and there are few trained linguists familiar with the language. An automatic speech recognition (ASR) system that can convert Miyako audio to text opens up new possibilities:

• Preservation: Create digital archives of oral histories before they are lost.
• Education: Generate subtitles for language lessons and authentic material.
• Research: Speed up linguistic analysis of recordings.
• Accessibility: Provide captions for community events.

Without ASR, these tasks remain out of reach for most communities.

The Challenges of Transcribing Miyako

Miyako presents several hurdles for ASR. First, it is a low-resource language: there is very little transcribed speech data available for training. Most global ASR systems, such as Google Speech-to-Text, Amazon Transcribe, or Whisper, do not support Miyako at all.

Second, the phonology is complex. Miyako distinguishes long and short vowels, a feature that can change meaning (e.g., pana 'flower' vs. pana: 'nose'). It also has a pitch accent system and glottal stops, which are rare in the languages that dominate ASR training data.

Third, dialectal variation is significant. The six main dialects (Ikema, Irabu, Kurima, Karimata, Nagahama, and central Miyako) differ in pronunciation, vocabulary, and even basic grammar. A model trained on one dialect may fail on another.

How Speechyou Tackles These Challenges

Speechyou's Miyako speech-to-text engine is built specifically for this language. We use transfer learning from related Japonic languages to bootstrap the model, then fine-tune on a hand-annotated corpus of Miyako speech. The system offers separate dialect profiles, allowing users to select the variety they are transcribing.

Acoustic modeling is adapted to handle vowel length, glottal stops, and pitch accent. The output is in the Latin script (Myaakufutsu), which is the most commonly used writing system for the language today. Users can export transcriptions as SRT or VTT subtitles, making it easy to add captions to videos or create study materials.

Use Cases in Action

From preserving the wisdom of elders to helping a new generation learn their heritage language, Speechyou is already being used in several ways:

• Oral history projects: Local organizations transcribe interviews with Miyako speakers to create searchable archives.
• Language classes: Teachers upload recordings of native speakers and generate subtitles for learners to follow along.
• Documentary filmmaking: Filmmakers add SRT subtitles to documentaries about Miyako culture, expanding their audience.
• Linguistic fieldwork: Researchers transcribe field recordings in real time, saving hours of manual work.

Conclusion

Miyako is a language with a rich history and a fragile future. Accurate speech-to-text is a powerful tool for keeping it alive. Speechyou is proud to support Miyako transcription, helping to ensure that every word, every story, and every song is captured for generations to come. Try it today and start transcribing your Miyako audio into text.