Speech Recognition
NVIDIA Trains Nemotron Speech Model on Saudi Audio Dataset to Cut Transcription Errors
NVIDIA used SDAIA’s Sada dataset to train its Nemotron 3.5 speech model, cutting Saudi dialect transcription error rates nearly in half.
SSICT News ·
Latest News
Cohere Launches Open-Source Speech Recognition Model for Arabic Dialects
Cohere has released a new open-source speech recognition model designed to transcribe 30 Arabic dialects and bilingual speech for enterprise applications.
OpenAI Launches Three Realtime Voice Models for Developers
OpenAI released three realtime voice models for developers: GPT-Realtime-2 for voice reasoning, GPT-Realtime-Translate for live multilingual conversations, and GPT-Realtime-Whisper for streaming transcription.