Speech to Text Working Group
Description
This WG handles the process of converting raw audios into standard audios of small chunks with corresponding verified transcript to create a training data that will be used for training the speech to text model. The model can be trained for general audios and for specific speakers as well.
Download & Documentation
Access the project’s resources:
- PRD: Custom STT Model PRD
- GitHub Repo:
- stt-audio-spliter
- stt-prepare-training-data
- stt-custom-model-finetuning
- stt-wav2vec2-finetune
- stt-whisper-finetune
- stt-model-evaluation - Documentation:
- Custom Speech to Text
- Time and Cost Estimation - Download Links:
- stt-general-model
- stt-multi-dialect-model
- Situ Rinpoche stt model
- Dilgo Kyentse Rinpoche stt model
Joining and Participating
To become a WG member, you must be registered on forum.openpecha.org. Members may be added or removed during regular WG meetings with quorum. Each membership change includes a reason, selected from the list below.
Reasons to Add a Member:
- Participation in WG meetings, events, or activities
- Contributions to WG-related projects
- Self-nomination during a WG meeting
Reasons to Remove a Member:
- No participation for over 3 months
- Violation of the OpenPecha Code of Conduct
- Personal request for removal
These are general guidelines; the WG may consider specific situations.
Communication Channels
- Discussion: OpenPecha Discord Server
- Documentation & Archives: OpenPecha Forum
- Meeting Schedules & Notifications: Sent via Google Calendar and Gmail mailing list
Meetings
Meeting invites are shared through Google Calendar. Please subscribe to the Gmail list to receive updates.
Working Group Members & Contacts
Name | Discord | GitHub | Role | |
---|---|---|---|---|
Ganga Gyatso | ganga@esukhia.org | Ganga01 | Ganga | WG Lead |
Gade | gade@pecha.org | Gade | gade | Annotator Lead |
David Yeshe Nyima | David N | Garchen Rinpoche stt representative |
What We’re Working On
We maintain a task board on GitHub. Check it to see ongoing work and ways to contribute: