Open Models & Projects
Models I have built and released openly through Modular on Hugging Face. Each model card documents its data, licence, results and limitations.
Speech · TTS
Dynamic Accent
Code-switching text-to-speech for English, Nigerian Pidgin, Igbo, Yoruba and Hausa, mixed inside one sentence in one consistent voice.
Speech · TTS
Dynamic Accent Pro v1.1
Experimental prompt-controlled version of Dynamic Accent: a plain-English style instruction steers pace and pausing.
Speech · ASR
Dynamic ASR
Speech recognition for Nigerian English, Pidgin, Igbo, Yoruba and Hausa, including speech that switches languages mid-sentence. A 1B-parameter fine-tune on 187 hours of licensed speech.
Speech · Voice agent
Dynamic Accent Duplex
Experimental full-duplex voice agent that answers in the same language mix you speak, built from permissively licensed components.
NLP · Embeddings
Native Bird
Sentence embedding model for Nigerian languages, for search, retrieval and similarity.
Multimodal · Retrieval
Native-Bird2
Image–text retrieval for Igbo, Hausa, Yoruba and English in one shared embedding space.
Multimodal · VLM
AtlasVision
Gives N-ATLaS, Nigeria's multilingual Llama-3-8B, the ability to see, via a SigLIP2 encoder and a trained projector.
Time series
Modular-TS-2.4M
A 2.4M-parameter time-series foundation model pretrained from scratch: zero-shot probabilistic forecasts that run on a CPU or phone.
Protein engineering
StaMP
Fast structure-conditioned predictor of single-point mutation stability (ΔΔG); scans every substitution of a protein in seconds on CPU.
3D vision
Polymorphic
Compute-efficient video-to-3D research combining AnySplat initialisation with Fourier Splatting.