אנחנו שמחים להזמין אתכם לאירוע MDLI – Speech Talks, שנעשה בשיתוף עם NICE. במהלך האירוע, שלושה דוברים מחברות מובילות יציגו התפתחויות מרתקות בתחום ה-Voice ועיבוד הסאונד: מזיהוי תרמיות באמצעות מדדים ביומטריים בקול, לשימוש בטרנספורמרים בתחום, ועד פרקטיקות לבניית הפייפליין.
האירוע יתקיים באופן מקוון ב-16.11.2021 ב-18:00-19:30. לרישום לאירוע לחצו כאן.
How NICE uses AI-based voice biometrics to prevent fraud in the customer service
Roman Frenkel, Expert Speech Researcher, NICE
Call centers are a prime target for fraudulent activity. Statistics show that contact centers worldwide handle more than 100 billion calls every month, and 1 in 1,200 calls is a fraud attempt. Luckily for us advancements in artificial intelligence, machine learning and voice biometrics have come together in modern platforms to enable real time authentication and fraud detection with astonishing accuracy. In this talk we will describe how NICE uses artificial intelligence and machine learning to fight fraud. We will present our holistic approach to fraud attacks detection both in real-time and offline using state-of-the-art voice biometrics technology which identifies a person by their unique vocal attributes.
How close are we to GPT-like zero/few-shot capabilities at speech processing?
Yoav Ramon, AI Lead, Hi Auto
In recent years, we have seen growing usage of Transformer models in the NLP community until almost all SOTA algorithms are incorporating those models in some way or another.
The transformers are good at getting outstanding results on existing benchmarks and by getting exceptional results even with only small amounts of data. This property of those models makes them highly useful for startups and advanced NLP research.
In the past year, we saw articles popping out promising the same things but in the domain of Speech/Audio processing like Wav2Vec or HuBERT.
In the lecture, we will ask if those works will have the same impact on the audio processing world and whether Speech Recognition/Speaker identification models will be based solely on transformers one day.
The lecture will also give practical tools on how to play and use those models and some internal information from early adopters of those models in the industry.
Building state-of-the-art voice pipeline
Dr. Jackie Assa, VoiceML Research, Snap
In this talk, We'll go over some of the different networks in building a high-scale state-of-the-art voice pipeline including ASR, TTS, NLU and Conversation management. I will describe some of the tradeoffs and lessons learned during our transition over the last year, from call center focused startup, to the voiceML group, in charge of voice services in Snap.
The talk will highlight some of the challenges and potential solutions for anyone who is interested in implementing conversational AI components