Limbu Speech to Text: Preserving a Tonally Rich Language with AI

Limbu, also known as Yakthungpan, is a Tibeto-Burman language spoken by the Limbu people in eastern Nepal, Sikkim, and Darjeeling. With approximately 700,000 speakers, it is one of the major languages of the Kiranti group. The language is written in both the Sirijanga script and Devanagari, though Devanagari is more widely used in digital communication and education. Limbu is a tonal language, featuring at least two contrastive tones (high and low) that can completely change the meaning of a word. For example, the syllable ta with a high tone means 'to see', while with a low tone it means 'to cut'.

Why Accurate Limbu Speech-to-Text Matters

Preserving Limbu in the digital world requires tools that can handle its unique phonology. Oral traditions, such as the Mundhum (sacred narratives) and shamanic chants, are passed down verbally and are at risk of being lost. Accurate speech-to-text allows these recordings to be transcribed into searchable text, archived, and studied. For community radio stations and YouTube channels in Limbu, automatic subtitles can greatly expand reach. Additionally, Limbu is a minority language in both Nepal and India, and digital tools help empower speakers to use their language in education and governance.

Specific Transcription Challenges

• Tonal Contrasts: The ASR model must distinguish pitch patterns. Speechyou uses pitch-based features and a tone-aware decoder.
• Limited Data: Few publicly available Limbu speech corpora exist. Speechyou employs transfer learning from related languages (e.g., Nepali, which shares some phonetic space).
• Devanagari Orthography: Limbu uses Devanagari but with some adaptations, such as the use of the chandrabindu for nasalization and special conjunct forms. Our model is trained on a custom lexicon of Limbu Devanagari.
• Dialectal Variation: The three main dialects (Panthare, Phedape, Tamarkhole) differ in vowel quality and some lexical items. Speechyou's training data includes examples from all three.

Use Cases in Detail

• Podcasts & Radio: Transcribe Limbu-language programs for show notes, searchability, and accessibility.
• Subtitles for Films: Generate SRT/VTT files for Limbu movies and documentaries, helping the diaspora follow along.
• Academic Research: Linguists can transcribe field recordings for phonetic analysis and language documentation.
• Education: Create transcripts of Limbu lessons for online learning platforms.
• Accessibility: Provide real-time captions for live events and meetings in Limbu-speaking communities.
• Oral History Preservation: Digitize elderly speakers' stories to create a permanent written record.

How Speechyou Helps

Speechyou offers a simple web interface where you upload your Limbu audio or video file and receive a transcription in Devanagari within minutes. The model is optimized for tonal accuracy and can handle background noise common in field recordings. You can also export subtitles in SRT or VTT format. The Solo plan includes unlimited transcription, making it affordable for community projects and independent researchers. Unlike most commercial ASR services, Speechyou actively supports Limbu and welcomes feedback to improve the model.

Getting Started

To transcribe your first Limbu audio, simply sign up for Speechyou, select 'Limbu (Devanagari)' as the language, and upload your file. The system will process it and return a text file with timestamps. For best results, use a clean recording with minimal background noise and a single speaker. If you have multiple speakers, try to separate them or note the change in the audio. With Speechyou, you can help keep the Limbu language alive in the digital age.