Using AI to get relevant KPIs for Audio Quality
summary
Good metrics include actual algorithms and results not buzzwords Here the claims are concrete: SpeechQ uses Whisper transcripts to score intelligibility by comparing ASR text to reference text, ASQ-ViT is a CNN backbone plus visual transformer that classifies spectrograms from only the degraded sample, and DNS MOS (Microsoft) is used for nonreference cases while POLQA and ViSQOL remain full reference. Training uses balanced data, end to end sender→receiver tests, and the models report very high correlation with subjective MOS and handle enhancement that breaks full reference scores
AI-written summary · about
related talks
05mRTC @Scale, Audio ML Session
21mDeveloping Machine Learning Based Speech Enhancement Models for Teams and Skype
26mObjectionable Uses of Objective Quality Metrics
27mUsing GenAI with Audio Data for Driving Better Customer Experience
13mVOICE Summit 2019 | VED Talks | Alexandros Potamianos | July 23
19mLLM-Enhanced Multimodal AI: Revolutionizing Audio/Video Interaction
20mLive Q&A 3 with Speakers
14mAI for a Billion Voices: Taking AI where there is Maximum Impact