Open Nepali ASR models and benchmark by Blackfire A.I. Free for research and non-commercial use.
AI & ML interests
Building open-source speech and language technology for Nepali and other underrepresented languages.
Recent Activity
Organization Card
Speech technology for Nepali, built in Lalitpur, Nepal.
Open models: Nepali speech recognition (Nepali ASR)
| Model | What it does |
|---|---|
| nepali-asr-xlsr-300m | Nepali speech to Devanagari text. 12.0% character error on FLEURS Nepali, the lowest of the public Nepali models we benchmarked. Runs on CPU. Nepali only. |
| nepali-asr-aud8-325m | Nepali speech to punctuated text. Handles occasional English words. Needs a GPU. |
Try it in your browser: Open in Google Colab. Upload a Nepali recording, get Devanagari text.
Collection: Nepali speech recognition
Both models are free for research and non-commercial use. For commercial use, contact hello@blackfire.com.np. Code, benchmarks and limitations: github.com/BlackfireAI
Products
- Aawazz: Nepali and English text-to-speech, voice cloning and dubbing
- LipiSub: video transcription, translation and subtitles
- Tipni: meeting notes for Nepali, English and mixed-language meetings
Contact: hello@blackfire.com.np
datasets 0
None public yet