Salesforce researchers explain how voice AI works: an acoustic model converts spoken sounds into characters, while a language model adds grammar and meaning. They're working toward human-level accuracy and predict future systems will be more intuitive and reliable.
Speech recognition is the field of trying to map spoken words into written words. That sounds very simple, but it actually enables a lot of powerful applications and personal assistants and personal conversational agents.
I think our research benefits average person. Now, you don't need to have a technical background to be interacting with devices. And if you have some disability or some other inconvenience, you can communicate directly with your voice.
Speech recognition system consists of two models. The first is acoustic model. The second second is language model. In the acoustic model, the computer will map pronunciation to alphabets and characters. In the language model, this computer will look into the grammar and semantics of the language. When we combine the two models, then we can predict the most likely text transcripts of the speech. It's exciting to be part of this effort to improve speech recognition performance to the human level. And hopefully.
Beyond that, I think the.
Future of AI for voice technology will be more accurate, more intuitive, more robust.
What's exciting about Salesforce Research is that we see the immediate impact on our customers and the products that we have. And that, I think, is part of what makes Salesforce Research so exciting.
