Track: Keynote VoiceTech |
| What analyzing call center conversations can teach us about good conversational AI |
| Speech technology is finding its way into many aspects of our lives. Some of us welcome it by bringing voice assistants into our homes, cars and pockets while others are forced into it, having to go through automated gatekeepers to reach company representatives. But all of us have experienced the utter frustration of speech tech that doesn’t work right. Getting speech technology to work well is hard. Natural human communication is spontaneous, fragmented, and disorganized while technology wants to be predictable, structured, and organized. Making good speech technology requires that we think about the humans as well as the algorithms. This keynote will take lessons learned from analyzing millions of real human conversations to provide insights into how we might improve our speech technologies. We’ll focus on three things that speech technology needs to get right; listening, recognizing, and understanding, and we’ll talk about the techniques we’ve used to improve each. This discussion will describe a speech analytics architecture that utilizes digital signal processing, distributional semantics, and prosodics to support better recognition, mitigate bias, and reliably model intent from long form two-party conversations. |
|
| Presentation Video |
| Presentation Notes |
| TAN-What-Call-Centers-Can-Teach-Us-About-Good-Conversational-AI-updated.pdf |
