Research & Innovation
Infrastructure Scientifique pour les Technologie Linguistiques Africaines
KivuLingua AI développe des systèmes avancés ASR et TTS utilisant des frameworks open-source: Whisper, wav2vec 2.0, MMS (Massively Multilingual Speech), SpeechBrain et Coqui TTS. Nous évaluons les performances via Word Error Rate (WER), Character Error Rate (CER), Mean Opinion Score (MOS). Notre recherche priorise la collecte éthique, la propriété communautaire et l'intégration avec Hugging Face et Mozilla Common Voice.
In Progress
Multilingual Speech Recognition for Low-Resource Bantu Languages using Whisper and wav2vec 2.0
Muhigiri Ashuza, CIRUZA Alain, KivuLingua Research Team
In Progress
Community-Driven Corpus Creation for Bantu Language Speech AI: Methodology and Ethical Frameworks
KivuLingua Project & Local Linguistic Communities
Planned
Benchmarking ASR Performance: Word Error Rate and Character Error Rate Analysis for East Congolese Bantu Languages
KivuLingua Technical Team
Planned
Neural Text-to-Speech Synthesis for Endangered Bantu Languages using SpeechBrain and Coqui TTS
CIRUZA Alain, Technical Development Team