Hey everyone,

This should be even bigger announcement. I wanted to share a project I worked on this summer with two interns. We tackled a problem that tech giants like Google, OpenAI (Whisper), and Meta have effectively ignored.

🚫 The Problem

In Cyprus, as you know, the official language is Standard Greek, but in daily life, most people speak the Cypriot dialect.

If you try to use AI speech-to-text apps or AI voice assistants here, they fail hard. Big models treat the dialect as "noise" or "bad Greek." Even Meta’s massive 1,600-language model doesn’t support it.

💡 The Plan vs. The Reality

We thought this would be a standard fine-tuning job:

1. Grab a Greek wav2vec model.
2. Download a dataset.
3. Burn some GPU credits.
4. Profit.

The Reality Check: There were NO datasets. Existing research data was either lost, broken, or locked behind a €35,000 paywall.

🛠 The Solution (The Hard Way)

Since data = AI, we had to build the entire pipeline from scratch. Here is what we did:

• Getting data: We downloaded hours of Cypriot TV, radio, and podcasts.
• Crowdsourcing: We built a platform (voiceofcyprus.org) to gather real translations from locals.
• Messy Data: We used every NLP trick to teach the model context using very imperfect audio data.

We’ve open-sourced everything:

https://huggingface.co/datasets/Elormiden/RIK_Cypriot_Collection_Dataset
https://huggingface.co/datasets/Elormiden/RIK_Cypriot_News_Dataset

🚀 The Result

In just 6 weeks and with only $150 in GPU creditswe trained a small model that actually understands the Cypriot dialect!
https://huggingface.co/Elormiden/bert-base-cypriot-greek

It isn’t production-perfect yet, but it is the first working pipeline. We proved that the task is solvable without a Big Tech budget. This provides a foundation for researchers and local devs to finally build voice AI for the island that can understand local people.

The Takeaway: You can have the best architecture in the world, but No Data = No AI

Check out the full breakdown of the project here: https://youtu.be/zN_FMIWRSLA

And for the Greek speakers, here is an AI-translated version of the explanation: https://youtu.be/hcoXFNVP6L4