When you talk to an AI character, a small chain of steps happens in the blink of an eye. Knowing the chain helps you get better results and understand why some habits make conversations smoother.
Step one: listening
Your microphone turns sound into a stream of tiny audio pieces. The app watches for the moment you start speaking and the moment you finish, so it knows when your turn is complete.
Clear audio helps here. A headset microphone close to your mouth gives the cleanest signal.
Step two: turning speech into text
A speech recognition model converts your audio into words. Names, places and unusual terms are easiest when spoken at a steady pace, and you can always repeat a word if you notice it was misunderstood.
Step three: writing the reply
A language model receives your words along with the character's personality description and the recent conversation. It then writes the reply in the character's voice. Characters with a clear personality and a clear goal produce focused replies.
This is why your opening line matters. Telling the character what you want, such as practice, advice or a story, steers every reply that follows.
Step four: speaking
A voice model turns the reply into audio, and the sentences begin playing as soon as they are ready. Short, natural turns keep the conversation lively, because the first sentence can start while later ones are still being prepared.
Tips that help every step
Speak at a comfortable pace and finish your thought before pausing. Use a quiet space. Keep headphones on so the voice stays out of the microphone. Ask for a short answer when you want a quick reply, and ask for more detail when you want depth.
If you enjoy the technical side, Lee Penkman wrote a developer tutorial on building a voice call agent step by step on How.nz, which goes into streaming, silence detection and interruptions.
Why short turns feel natural
Human conversations move in short exchanges, and AI calls follow the same pattern. A sentence or two from you and a sentence or two from the character keeps energy high and gives each side room to respond. Longer monologues are welcome too, and the character will reply once you pause.
If you would like the character to wait while you think, say so at the start, for example: "I might pause a little while I figure out my words." Clear instructions at the beginning shape the whole call.
Choosing the right voice for the job
Different characters suit different purposes. A calm helper is wonderful for planning and reflection. A dramatic narrator brings stories to life. A cheerful companion is great for casual chats. Browse the character list, read the descriptions, and try a couple of calls to find the voice that fits your moment.
Because each character carries its own personality, the same question can lead to delightfully different answers. That variety is one of the joys of calling instead of typing, and it invites you to keep exploring.
Takeaway: Listening, transcription, a character reply and a spoken voice work together, and a quiet room plus clear goals make each step shine.
Characters to try
Read the developer tutorial on How.nz, or pick a character and try a call.