Keiyo Speech to Text: Preserving a Kalenjin Language with AI

The Keiyo Language and Its Speakers

Keiyo (also known as Elgeyo) is a Southern Nilotic language of the Kalenjin branch, spoken primarily in the Rift Valley region of Kenya. With an estimated 300,000 speakers, it serves as a vital medium for daily communication, cultural ceremonies, and oral traditions. The language is written using a Latin-based orthography developed by missionaries and adapted by local language committees. Despite its importance, digital resources for Keiyo are scarce, and few speech recognition tools support it.

Why Keiyo Speech to Text Matters

Accurate speech-to-text for Keiyo opens up new possibilities for the community:

• Preserving oral heritage: Transcribe interviews, folk tales, and proverbs before they are lost.
• Education: Provide text versions of Keiyo lessons for students and literacy programs.
• Media accessibility: Add subtitles to Keiyo videos for the deaf and for wider audiences.
• Linguistic research: Create searchable corpora for study of Kalenjin languages.
• Local radio and church: Convert sermons and announcements to text for distribution.

Transcription Challenges in Keiyo

Keiyo presents several challenges for automatic speech recognition:

Tonal system: Like many Nilotic languages, Keiyo uses tone to distinguish words. For example, /kèːp/ means "to count" while /kéːp/ means "to tie". A speech engine must detect pitch patterns to avoid errors.

Phonetic complexity: The language features prenasalized stops (e.g., /nd/, /mb/) and labiovelar consonants (e.g., /ŋʷ/), which are uncommon globally.

Dialectal variation: Keiyo itself varies between regions, and speakers of Marakwet or Sengwer may use different lexemes and pronunciations. Speechyou’s model accommodates these variations.

Scarce data: With few hours of transcribed speech available, building a robust ASR from scratch is difficult. Speechyou overcomes this through transfer learning and continuous model updates based on user feedback.

Use Cases in Detail

• Oral History Preservation: The Keiyo community has a rich tradition of storytelling. By recording elders and transcribing their narratives, families and researchers can archive these stories. Speechyou enables batch processing of long recordings, making large-scale documentation feasible.
• Educational Materials: Schools in Keiyo-speaking areas can use speech-to-text to create written versions of lectures or to help students practice reading in Keiyo. The subtitles feature also helps produce dual-language content.
• Church and Religious Content: Many churches in the region conduct services in Keiyo. Sermons can be transcribed for study groups or uploaded as subtitled videos for remote communities.
• Community Media: Local radio stations and social media influencers create content in Keiyo. Adding subtitles expands reach to non-speakers and improves accessibility.
• Language Activism: Activists can use Speechyou to transcribe speeches and meetings, creating written records that support language planning and policy.

How Speechyou Handles Keiyo

Speechyou’s AI is trained specifically on Keiyo data, including recordings from multiple age groups and regions. The model is tuned to recognize the tonal and phonetic patterns of the language. Users simply upload audio or video, select Keiyo as the language, and receive accurate transcripts. The tool also supports real-time transcription and exports in SRT, VTT, TXT, and other formats, making it a comprehensive solution for Keiyo speech-to-text needs.

Future Prospects

With ongoing improvements and community contributions, Speechyou aims to increase accuracy further and expand support for other Kalenjin languages. By providing accessible transcription for Keiyo, we hope to empower speakers to preserve their language in the digital age.