Cobus Greyling joins us on our podcast to talk conversational AI tooling, and how to make conversational AI work for large enterprises.
Presented by Deepgram
Deepgram is a Speech Company whose goal is to have every voice heard and understood. We have revolutionized speech-to-text (STT) with an End-to-End Deep Learning platform. This AI architectural advantage means you don’t have to compromise on speed, accuracy, scalability, or cost to build the next big idea in voice. Our easy-to-use SDKs and APIs allow developers to quickly test and embed our STT solution into their voice products. For more information, visit:
<
How to make conversational AI work
Cobus Greyling led Vodacom’s Tobi implementation in South Africa, and is a well-known entity in the conversational AI space. His consistent analysis of leading conversational AI platforms on Medium is a must read for anyone serious about CAI. After leaving Vodacom and having the public speaking shackles released, Cobus joins me to share his take on how to make conversational AI work for large enterprises.
00:00 Introduction and presenting Deepgram
Register for ‘The irony of big tech: why your speech recognition is probably a load of tosh’: https://vux.la/3ySe8n8
02:40 Welcome Cobus Greyling
07:10 Writing on conversational AI
08:20 The irony of big tech
14:16 Two problems for orgs delivering voice experiences
19:25 The emergence of AI verticles
22:55 5 categories of conversational AI
31:20 Open NLU pipeline
34:05 The rise of voicebots
39:00 4 metrics to measure voicebot success
43:15 Transcript reviews: 1 hour a day
44:40 Making voicebot open and more verbose
45:45 Establishing themes instead of intents
47:55 Disambiguation
51:30 Scale and fine tuning
54:20 The commercial application of large language models
55:46 Tuning speech recognition with custom acoustic models
1:00:55 Intent deprecation and technology consolidation
1:06:28 Barge in
1:10:27 Audio design
1:11:35 End of speech detection
1:13:00 “Hello” and easter eggs
1:14:50 Outro
Links
Visit Cobus Greylings website
Register for ‘The irony of big tech: why your speech recognition is probably a load of tosh‘