Hearing her is what turns a chat app into a companion. Most AI girlfriend apps now offer two voice features: voice notes, where a message she wrote is read aloud in her voice, and live calls, where you speak and she answers out loud in real time. This guide explains how both work, why they cost what they cost, and how much quality depends on your language, using measurements Lovora ran on its own voice engine.
Voice notes and live calls are different features
- A voice note is her text message turned into speech. It takes a few seconds and is the cheaper of the two.
- A live call chains three steps for every exchange: speech recognition turns your words into text, the language model writes her answer, and speech synthesis speaks it. The delay you hear is the sum of those three.
On Lovora, a voice note costs 1 credit and the first 2 are free with a free account. Calls cost 25 credits per minute and are metered on the seconds actually spoken, so silence and thinking time are not billed.
How a call works, step by step
- You hold the microphone button and speak.
- Speech recognition transcribes you. On Lovora this takes well under a second in all nine languages.
- The model writes her reply in character, in the language you are using.
- Speech synthesis produces her voice from a short reference clip that defines her timbre; on Lovora each voice in the picker is one of 24 references.
- Her audio plays, and the next turn begins.
Voice quality depends on your language
Speech engines are trained on uneven amounts of data per language, and it shows. Lovora measures quality by having her voice read sentences and transcribing them back with a speech recogniser: the share of words that come back different is the word error rate. Lower is better.
| Language | Measured error rate | What it sounds like |
|---|---|---|
| English, Italian, German, Portuguese | 0% | Clear and natural |
| Chinese (Simplified) | 0% per character | Clear |
| French | ≈ 4% | Clear; an occasional repeated word |
| Japanese | ≈ 10% per character | Mostly clear; the odd sentence misheard |
| Arabic | ≈ 11% | Understandable; a word occasionally mispronounced |
| Spanish | 0% to ≈ 20% | Varies with the sentence |
Measured in September 2026 with five sentences per language on the engine in production. Any app can publish figures like these; if an app claims natural voice in your language, ask it how it measured.
Why calls are billed per minute
Every second of her voice is generated on a GPU at the moment of the call, and recognition runs on your audio in parallel. That makes a minute of call far more expensive to serve than a minute of text chat, which is why almost every app meters calls or reserves them for paid tiers. Metering by speech rather than by connection time is the fairer model, because you are not paying for pauses.
What makes a call feel real
- Latency: under two or three seconds between your last word and her first one.
- A voice that matches her photos and personality.
- Memory that carries over from text: she should know in the call what you told her in chat.
- Staying in the same language, even if the app interface is in another.
What to check before paying for voice
- Whether voice notes and calls are both included, and how each is charged.
- Whether you can hear her voice before paying; on Lovora, every voice in the picker has a preview in each language.
- Quality in your language, tested during the free tier.
- What happens to the audio: whether recordings are stored.
Voice on Lovora, in short
Tap ▶ under any message to hear it in her voice, or start a call from the chat. A free account includes 2 voice notes; plans include 500 credits a month, which cover about 20 minutes of calls, or any mix with photos and videos. The 24-hour pass includes 150 credits, enough for about 6 minutes. Details are on the Voice notes and voice calls page.
FAQ
Can an AI girlfriend call you?
Yes. On apps with live calls, including Lovora, you start a call from the chat, speak, and she answers out loud in real time.
How much do AI girlfriend voice calls cost?
Usually a per-minute charge in credits or tokens, or an included allowance in higher tiers. On Lovora a call costs 25 credits per minute of speech; a plan's 500 monthly credits cover about 20 minutes.
Is AI voice good in languages other than English?
It depends on the engine. On Lovora's measurements, Italian, German, Portuguese and Chinese came back without errors, French with about 4%, and Japanese, Arabic and Spanish with more variation.
Are my calls recorded?
Check each app's privacy policy. The audio is processed to transcribe and answer you; look for whether it is stored afterwards.
Can I hear her voice before paying?
On Lovora, yes: each voice has a preview, and a free account includes 2 voice notes.
