AI companion voice calls: languages, credits and real costs

Research updated 26 October 2026. Documentation-based editorial guidance; these are proposed reader tests, not results of our paid hands-on testing.

Voice messages, spoken replies and live calls are different products

A service can offer a button that reads text aloud without supporting live conversation. Text-to-speech speaks a completed reply; voice messages let you record or receive short clips; real-time calls continually process your speech and generate responses. Video or animated-avatar calling may add a separate charge. Before buying, identify which interaction you actually need and whether it is included in your plan.

Nomi says paid subscribers can use its in-house and custom voices without a Nomi usage limit; an optional ElevenLabs integration follows the allowance of the connected ElevenLabs plan. Nomi’s documentation also cautions that voice is designed primarily for English and other languages can sound inconsistent. “Voice available” therefore does not mean natural German, French or Spanish pronunciation.

Documented examples that affect cost

Provider	Published voice condition	Buying implication
Nomi	Paid voice; in-house/custom voices described as unlimited	Verify the accent and language yourself before choosing a subscription
Kindroid	Subscriber audio credits; V3 costs 1.5× and V4 1.75× the V2 audio rate	More expressive voice can reduce available listening or call time
OurDream	Voice calls cost 50 Dreamcoins per minute under the published tariff	Ten call minutes consume 500 coins; 30 minutes consume 1,500
Candy AI	Paid tokens may be used for voice messages and beta voice calls	Unlimited text chat does not establish unlimited spoken interaction

Kindroid also documents an additional 400 audio credits per minute during active voice or video calls, alongside audio-generation usage. A call of approximately 10 minutes therefore incurs about 4,000 call-processing credits, plus the credits needed for speech generation and any selected live-avatar feature. That is a credit illustration, not a pound price. Plans and top-up credit prices must be checked at checkout.

For OurDream, 20 minutes × 50 = 1,000 Dreamcoins. Its published subscriber allocation is 1,000 monthly coins, so this hypothetical call habit alone could exhaust that recurring allowance; other paid actions would require extra coins. That does not mean 20 minutes is a fixed maximum, since balances may vary and coins can be purchased.

An honest five-part voice test

Recognition: read a sentence with an unfamiliar place name. Does the system transcribe or respond to the right words?
Pronunciation: ask it to say the place name back, plus a date and a number. Note clipped sounds and English-style pronunciation.
Turn-taking: pause mid-sentence, then continue. Does the system interrupt or wait naturally?
Latency: time three ordinary replies with a phone stopwatch. Record the device, connection type and whether responses were unusually long; do not claim a universal speed from one trial.
Cost: open the credit history before and after a five-minute call. Check whether listening, regeneration, live processing and video each create charges.

Test using non-sensitive sentences. A voice call may involve microphone access, temporary media processing and network transmission; review the provider’s actual storage and privacy practices before making intimate calls.

UK-specific purchase questions

If the interface shows a US-dollar tariff, do not publish it as a guaranteed sterling price. Confirm the GBP checkout amount, whether voice features work on web and mobile, and the payment channel used. If buying through Apple or Google, investigate that store’s subscription-management path rather than assuming the provider’s website cancels the store purchase. On shared devices, test with headphones and review microphone permissions.

Decision: Nomi may appeal to buyers who value ongoing spoken conversation and accept English-first voice performance. Kindroid gives clearer granular control but exposes several layers of audio-credit consumption. OurDream makes a per-minute coin calculation straightforward. Which voice is actually more natural cannot be concluded from vendor documentation; our editorial team has not performed comparative call tests.