I solve multilingual Natural Language Processing (NLP), Automatic Speech Recognition (ASR), and Text To Speech (TTS) problems. I also work on developing multilingual conversational AI systems. Through my journey, I had the opportunity to teach Machine Learning (ML) courses across multiple
universities, to deliver workshops, and participate in worldwide tech conferences.
Work & Experience
1. Train ASR models- Pre-process and normalize speech recognition datasets.- Apply transfer learning for Arabic and North-African dialects. - Manipulate Connectionist Temporal Classification (CTC) models: Jasper, Quartznet, and Citrenet models using NVIDIA NeMo package and Wav2Vec2 model using Fairseq and Hugging Face packages. 2. Train TTS models - Create and pre-process speech synthesis datasets. - Train Text to Speech models for Arabic and North-African dialects. - Manipulate Glowtts and Tacotron2 using the NVIDIA NeMo package. 3. Build Conversational Question Answering System - Create question-answering datasets. - Build Arabic and North-African dialect chatbots using the Rasa framework.
- Scrap and pre-preprocess social media text data. - Build a word2vec language model for the Tunisian dialect, and use the learned word embedding to showcase impressive similarity results. - Train an Attention-based model inspired by the Transformer architecture on the Multi-Turn Context Response Selection NLP task and show great prediction results.