🚀 Excited to Share my work on Speech-to-Text Systems

🚀 Excited to Share my work on Speech-to-Text Systems #LinTO ! 🎤

I’m thrilled to share a video showcasing the results of my recent work on developing a Speech-to-Text system for the Tunisian dialect. As a Research Engineer in #LinagoraLabs at LINAGORA, I’ve had the opportunity to dive deep into ASR (Automatic Speech Recognition) technologies, particularly using the Kaldi toolkit.

🔍 This model is especially unique because it can handle code-switching between Tunisian Arabic, French, and English, reflecting the way many people naturally speak in Tunisia.

💼 What I Did:

• Language Model Training: Crafted a robust language model to accurately capture the nuances of Tunisian Arabic and seamlessly switch between languages.
• Acoustic Model Training: Developed an acoustic model to ensure clear and precise recognition, even with the variations in pronunciation and dialect.
• Word Normalization: Implemented advanced word normalization techniques to standardize the output, making it more accurate and user-friendly.
• Data Augmentation: Enhanced the model’s robustness by using various data augmentation techniques, including voice conversion, integrating music data, and isolating vocals from music tracks to diversify the training data.

🎥 Check out the video to see the results and learn more about the work we’ve been doing at Linagora. I’m looking forward to your feedback and thoughts!

#ASR #SpeechRecognition #STT #SpeechToText #Kaldi #AcousticModel #LanguageModel #MachineLearning #DeepLearning #TunisianDialect #CodeSwitching #DataAugmentation #NLP #AI #Linagora #ResearchEngineer #ArtificialIntelligence