Celebrating the Successful Dagbani Speech Technology Training Workshop by the Dagbani Wikimedians User Group

by
Translate this post

On 28th June, 2025, the Dagbani Wikimedians User Group hosted a successful in-person training workshop as part of our project, “Training Dagbani ASR and TTS Models with Mozilla Common Voice.” This initiative was generously funded by the Mozilla Foundation to develop cutting-edge speech technologies for the Dagbani language. A major Gur language spoken in northern Ghana. 

The workshop brought together language activists, Wikimedians and community members to build local capacity around data collection, recording, and annotation for Automatic Speech Recognition (ASR) and Text-to-Speech (TTS) development.

The training covered various aspects of Mozilla Common Voice, including:

  • Platform Navigation: Participants were guided through the interface of the Common Voice platform, understanding its layout and functionalities.
  • Data Upload Procedures: Detailed instructions were provided on how to correctly format and upload Dagbani text data. This included understanding the required metadata and file specifications.
  • Quality Control and Validation: Emphasis was placed on the importance of data accuracy and quality. Members learned how to validate existing sentences and ensure the integrity of newly uploaded text.
  • Community Contribution Guidelines: The session also highlighted the broader impact of contributing to Mozilla Common Voice and the guidelines for collaborative efforts within the community.

The energy and commitment from all attendees reaffirm our shared mission to preserve and promote the Dagbani  language through open technology. 

Key Project Deliverables

This project has three primary goals:

  1. Fine-Tuning ASR Models
     Using open-source speech models like Wav2Vec2.0, we are applying fine-tuning techniques with Dagbani data from Mozilla Common Voice to create a high-performance ASR model. Our aim is to achieve a Word Error Rate (WER) below 25% while accurately recognizing Dagbani’s tonal features and dialectal variations.
  2. Developing TTS Models
     We are also building a TTS system that will convert Dagbani text into natural, intelligible speech. The target is a Mean Opinion Score (MOS) of 3.5+ for naturalness and tone accuracy, ensuring users hear Dagbani as it’s meant to be spoken.

Open Source Access and Community Engagement
All resulting models, datasets curated from the Dagbani Wikipedia, and tools will be released under an open-source license CC0. We will also present our findings in an upcoming virtual event organized by Mozilla for the global Common Voice community.

Why This Matters

Dagbani is a tonal language with unique phonetic and dialectal characteristics. Yet, like many African languages, it lacks robust technological tools. Through this initiative, we aim to bridge that gap  by enhancing language accessibility, supporting educational tools, and enabling voice-enabled AI solutions in Dagbani.

What’s Next

  • Publish ASR and TTS models for Dagbani under an open-source license
  • Contribute over 10,000 validated Dagbani sentences to the Common Voice platform
  • Collaborate with Mozilla and local communities for ongoing model improvements
  • Lay the groundwork for supporting other Gur languages through transfer learning

We extend our heartfelt thanks to the Mozilla Foundation for their support and belief in this project. Together, we are shaping a future where African languages thrive in the digital world.

Learn more or get involved:
🔗 Common Voice – Dagbani
 🔗 Dagbani Wikimedians User Group

Can you help us translate this article?

In order for this article to reach as many people as possible we would like your help. Can you translate this article to get the message out?