Voice bots are software that answers a phone call and handles it from start to end, with no person on the line. They are also called voicebots, AI voice agents or virtual agents. The caller talks normally, in full sentences, and the bot replies aloud, in a voice that sounds like a person's.
A voice bot repeats three steps for every sentence. First, speech recognition converts the caller's words into text. Second, a language model reads the text and decides what to do and what to reply. It can look up information in the company's systems, such as an order or an account. Third, text-to-speech software converts the reply into audio and plays it to the caller. These models are described in speech and voice AI.
The call itself reaches the bot over the internet, as VoIP (voice over IP), voice carried as data. A company usually rents the phone number and the call connection from a messaging and voice API company. It then points the number at the bot.
Here is a typical case. At 11 p.m., the internet stops working in a home in Kochi. The customer calls the broadband company's helpline and says, in Malayalam mixed with English, "My Wi-Fi has not worked since ten o'clock." The bot finds the customer's account from the calling number. It checks the network records and sees a cable cut in the area, already reported. It tells the customer that engineers are working on it and that service should return by 6 a.m. It offers to send an SMS when the line is back. The call takes about 90 seconds, and no agent is needed.
Two things make voice bots harder to build than chatbots. The reply must start within about a second, or the caller thinks the line has gone dead. And callers interrupt, change their minds and mix languages, as many Indians do. The bot must stop speaking when the caller starts, and still follow what was said.
When the bot cannot help, it passes the call to a person in the contact centre, together with a written summary. So the customer does not have to repeat the problem.
How are voice bots different from the old phone menu, which says "press 1 for billing"? That menu is called IVR, and it only understands key presses. A conversational IVR understands a spoken request, but mainly to send the call to the right place. A voice bot goes further and handles the whole request itself. A voice bot that also acts in the company's systems, for example to book a refund, is one kind of AI customer service agent. Before a company lets its bot answer real customers, it tests it with hundreds of imaginary callers, through voice agent testing.