Spaces:
Running
Running
| --- | |
| title: MauriAI | |
| emoji: 🎙️ | |
| colorFrom: green | |
| colorTo: gray | |
| sdk: static | |
| pinned: false | |
| <img src="https://huggingface.co/spaces/Emin009/HAKEM/resolve/main/mauriai_logo.png" alt="MauriAI" width="260"> | |
| # Speech AI from Mauritania | |
| MauriAI builds voice technologies with a focus on Hassaniya and multilingual speech applications. | |
| **Founder and researcher:** Mohamed Lamine Bewa · Mauritania | |
| ## HAKEM | |
| **Hear an item. Select it from your list.** | |
| HAKEM uses frozen speech models to select among supplied text candidates, subtracting a matched digital-silence score. It requires no additional task-specific weight training or recorded enrolment examples of the candidates. | |
| [Try the live HAKEM demo](https://huggingface.co/spaces/Emin009/HAKEM) · [Read the research preprint](https://doi.org/10.5281/zenodo.23326833) · [MauriAI website](https://mauriai.net) | |
| The demo compares candidate scoring with Whisper Tiny transcription followed by text matching, on the same user-supplied recording and list. It is an implementation preview, not an independent replication of the paper's benchmarks. | |
| **Research scope:** selection from a supplied catalogue containing the correct item. This is a preprint, not yet peer reviewed. It does not establish general transcription accuracy, spontaneous command understanding, or universal language performance. | |
| **Privacy:** Hassaniya research recordings and the private study catalogue are not distributed here. The demo uses visitors' own audio, processed on its host with temporary framework files. Please use non-confidential test recordings. | |