I will fine tune a mongolian speech recognition model
About this gig
Need speech-to-text that actually works for Mongolian or another low-resource language? I lead AIMOND VOICE, a Mongolian speech recognition project: 111,168 recordings (64.7 hours) collected, and word error rate (WER) reduced from 99.5% to 9.4% on a held-out evaluation set.
What I do:
- Evaluate your current model on your own audio and report WER with error analysis
- Prepare and clean your dataset (audio, transcripts, splits)
- Fine tune Whisper or a similar open model on your data
- Deliver the model, the evaluation report and a simple inference script
How I work: written communication only, clear milestones, reproducible results. Training runs on my own GPU server unless you require your own environment. Your data is used only for your project.
Please message me before ordering to confirm the language, data size and target accuracy.
Get to know Sergelen G
Mongolian AI, Feasibility and Research Consultant
- FromMongolia
- Member sinceNov 2026
Languages
English, Mongolian
FAQ
Do you use AI tools?
Yes, for data preparation and analysis where you allow it. Every result is checked by a person. If you do not want AI tools used on your data, tell me before ordering.
Can we talk by video call?
I work in writing: messages, documents and milestone reports. This keeps every decision on record.
Which languages do you support?
Mongolian is my specialty. I can also work with other low-resource languages if you provide transcribed audio.
