Text is fine. Voice is different
Reading 'I missed you' is one thing. Hearing her say it, with a pause before the words, lands differently. Voice is the feature that most changes how a companion feels — more than any model upgrade.
On WildNight, voice replies are generated from the same conversation as text: she says what she would have typed, in a voice matched to her character.
How it works technically
The pipeline is speech-to-text for your voice messages, the language model for her reply, and text-to-speech for her voice. Each step adds a little latency, so a voice reply takes a few seconds rather than being instant.
That's normal for every voice companion service right now. Anyone promising human-speed conversation with no delay is overselling.
What makes a voice good
The difference between pleasant and uncanny is prosody — pacing, emphasis, small hesitations. A flat, fast read of a long paragraph sounds like a GPS. A good companion voice keeps replies shorter and more spoken in style.
It also needs to fit the character. A shy anime companion and a confident realistic companion shouldn't sound identical, so voices are picked per personality rather than one-size-fits-all.
Try it sensibly
Start with text in the free guest preview, then test voice once you know you like her personality. Voice is a premium-feeling feature on any platform, and it pairs best with an account so she remembers previous conversations.
Use headphones. Seriously. It makes a bigger difference than any setting.