Semelai Speech to Text: Preserving an Endangered Aslian Language with AI

The Language and Its People

Semelai is an Austroasiatic language belonging to the Aslian branch, spoken by around 5,000 people in the tropical forests of Peninsular Malaysia. The majority live in small settlements along the Pahang and Bera rivers in Pahang state. Despite its small speaker population, Semelai possesses a complex phonological system and a deep cultural heritage, including oral epics and shamanistic traditions. Like many indigenous languages, it faces pressure from Malay, the national language. Speech-to-text technology offers a new avenue for documenting and revitalising Semelai.

Why Accurate Semelai Transcription Matters

For linguists, anthropologists, and the community itself, accurate speech recognition can revolutionise language preservation. Manual transcription of even a one-hour interview can take a full day. An AI that understands Semelai’s unique sound system — including prestopped nasals, breathy vowels, and final glottal stops — reduces this effort to minutes. This means more time can be spent analysing content rather than transcribing it.

The Specific Challenges of ASR for Semelai

Semelai offers several hurdles for automatic speech recognition:

• Prestopped nasals and plosives: Sounds like [ᵇm] and [ᵈn] are rare globally and not present in major training languages.
• Vowel length and breathiness: Contrasts such as /i/ vs /iː/ and breathy vowels require fine acoustic discrimination.
• Data scarcity: No existing large-scale speech corpus for Semelai. Most ASR vendors skip it entirely.
• Code-switching: Speakers frequently mix Malay words, so the system must handle bilingual input seamlessly.

Speechyou addresses these challenges head-on. Our model is trained on field recordings and supplemented with synthetic data from related languages of the Mon-Khmer family. We also allow users to upload domain-specific vocabulary.

Practical Use Cases in the Community

• Oral history documentation: Elders can tell stories in Semelai, and Speechyou generates written transcripts in the Latin orthography, creating a permanent record for future generations.
• Educational videos: Teachers produce subtitled lessons for children, helping them connect spoken Semelai with written form.
• Community media: Podcasts and Facebook Live streams automatically get captions, making content accessible to relatives who are more fluent in Malay.
• Healthcare communication: Health workers record instructions in Semelai and get accurate transcriptions for filing or translation.
• Linguistic research: Graduate students in anthropology can transcribe months of fieldwork data in a weekend.

How Speechyou Helps

Unlike generic tools that ignore Semelai, Speechyou is built to handle its specific phonology and orthography. The platform supports audio and video uploads in common formats and outputs SRT and VTT subtitle files. It runs in the cloud — no specialized hardware needed. And with the Solo plan, you get unlimited transcription minutes, so cost is never a barrier.

The Future of Semelai Transcription

As more Semelai speakers use the tool, the model improves via active learning. Each transcription contributes to a growing corpus that makes future outputs even more accurate. This virtuous cycle means that Semelai — a language once invisible to technology — can now share the digital stage with the world’s major languages.