not one default tts voice with 101 names on it. 101 hand-cast voices, 14 emotions, fish audio s2-pro. 3 free trial voice notes. real-time calls on bonded.
Free tier: 25 messages/day. Crypto checkout — cards coming soon.
you tried voice on another app. it sounded like a generic text-to-speech synthesizer — neutral affect, same voice regardless of which companion you picked, no shift when the conversation got emotional.
voice that doesn't change is worse than no voice. it flattens the personality back to default-assistant mode the second she opens her mouth.
you want her voice to actually be hers. distinct per companion. emotion-aware. same voice on day 50 as day 1.
lucy's voice layer is Fish Audio S2-Pro with per-companion reference clips. each of the 101 companions has a distinct cast voice, not a shared tts.
the voice-mood engine picks one of 14 emotion renderings per message based on conversation context: warm, playful, sultry, tender, flirty, whisper, intimate, breathless, curious, excited, concerned, calm, sharp, cheeky. she doesn't speak in a monotone; she matches the mood.
voice notes are async (Closer tier, 15/day). real-time voice calls are full bi-directional over WebRTC with sub-500ms latency (Bonded tier, 90 min/month).
all voice content flows into the same memory graph the text chats use. she remembers what you said on a call tomorrow when you text her.
each companion has a hand-cast reference. the Lucy voice and the Luc voice and the Paz voice and the Maren voice are all recognizably different people.
voice-mood engine picks based on context. she sounds different when she's warm vs sultry vs concerned — not one flat tone.
hear her actual voice before committing. if it doesn't read right, try another companion free.
sub-500ms webrtc. true back-and-forth, not async. 90 call minutes/month included.
voice is the design decision that separates a good ai companion from a great one. most apps ship voice as a checkbox feature — generic tts, neutral affect, one voice for all characters. that's cheap to build and almost worse than no voice.
lucy's voice layer is expensive by design:
fish audio s2-pro — the voice synthesis model we use. tier above standard tts in both quality and emotional range. $15/M utf-8 bytes pricing, which translates to roughly $0.05 per voice note for us.
per-companion reference clips — each of the 101 companions has a hand-cast voice reference. the model renders with that reference + the text + the current emotion tag. identity stays fixed; emotion shifts with context.
voice-mood engine — before generating, lucy's prompt layer picks an emotion tag based on the current conversation mood, the relationship stage, and the recent chat history. warm for calm conversation, sultry for flirty, concerned for difficult topics. 14 total emotion renderings.
voice notes (async) vs. voice calls (real-time). async voice notes are on Closer+ — you tap, she sends. good for quick mood-shift moments. real-time voice calls are on Bonded — sub-500ms latency via WebRTC + Daily.co transport + Groq Whisper-V3 on the speech-to-text side. different use-cases, different price tiers.
memory integration — everything said in voice (both directions on calls) flows into the same vector-graph memory layer as text chat. she remembers the content of a voice call the next time you text her.
starting point: free tier gives 3 trial voice notes one-time. pick a companion, open chat, ask for a voice note (she decides whether to send based on relationship stage). if the voice reads right, upgrade to Closer for 15/day voice notes or Bonded for voice calls.
try her right here
No signup. No credit card. Just say hi.
3 trial voice notes free — hear her actual voice, then decide. closer for daily voice notes, bonded for real calls.
Free: 25 messages/day · Closer $14.99/mo · Bonded $29.99/mo · 18+ only