VBD 
Thanks to relentless efforts in driving innovation, the research work of VinBigdata’s Natural Language and Speech Processing team has been officially accepted for presentation at INTERSPEECH 2024—the world’s largest and most comprehensive conference on speech processing technology. The event organized by the International Speech Communication Association (ISCA) will take place from September 1-5, 2024, on the island of Kos, Greece.
Công trình nghiên cứu của VinBigdata với đề tài “Improving Speech Recognition with Prompt-based Contextualized ASR and LLM-based Re-predictor”, nghiên cứu đã làm nổi bật những những đột phá mới nhất trong lĩnh vực hệ thống nhận dạng tiếng nói nói tự động (Automatic Speech Recognition – ASR), hướng tới mục tiêu chuyển đổi giọng nói thành văn bản với độ chính xác vượt trội hơn.
The research “Improving Speech Recognition with Prompt-based Contextualized ASR and LLM-based Re-predictor” highlights groundbreaking advancements in Automatic Speech Recognition (ASR) systems. It focuses on enhancing speech-to-text conversion accuracy with cutting-edge AI methodologies.
VinBigdata introduces a novel hybrid architecture integrating generative AI to address existing ASR limitations, particularly in challenging conditions such as noisy environments, regional accents, lack of contextual information, and recognition of rare words (proper names, loanwords, etc.). Experimental results demonstrate the solution’s effectiveness, achieving an average 30% reduction in Word Error Rate (WER) compared to conventional ASR systems
This achievement underscores VinBigdata’s continuous dedication to enhancing product quality and delivering cutting-edge technological solutions that make a real-world impact.
