Towards Large-scale Speech Analytics Systems
summary
Claims a production system that audits telemarketing sales at scale by combining Whisper transcripts, voice activity detection, diarization and LLMs with SQL/NoSQL metadata to monitor 30k sales per month and raise audited coverage from ~1% toward full coverage. It replaces a $1M human audit cost with parallel cloud orchestration, n-to-1 and one-to-m approaches for mapping script to speech, Levenshtein for fuzzy names, word-level alignment for numeric accuracy, and projects 6.7 million in value capture displayed in PowerBI.
AI-written summary · about
chapters (5)
- 0:00Speech analytics use case
- 5:06Solution technical overview
- 10:54Audio transcription pipeline
- 17:20Data workflow infrastructure
- 20:33Future enhancements
related talks
32mHarnessing Large Language Models for Enhanced Customer Experience
23mContact Center Voice AI: Low-Latency Intelligence Extraction from Messy Audio Streams
35mTricks Learned from Scaling WhisperSpeech Models to 80k+ Hours of Speech
26mEnabling the next generation in voice technology
17mKeynote: NLP Trends and Investments
19mLLM-Enhanced Multimodal AI: Revolutionizing Audio/Video Interaction
27mUsing GenAI with Audio Data for Driving Better Customer Experience
11mScaling Conversational AI in Production