AI companion voice chat technology
Built for fast, private-feeling AI voice chemistry.
iTsChat is a web-first fictional AI companion product with a protected voice pipeline built around near-instant speech-to-text feedback, character-shaped AI replies, and spoken responses that keep the conversation moving.
Live beta signal
Near-instant xAI STT has been observed in live beta testing, including low-signal 4G use.
We describe this as observed beta behavior, not a guaranteed latency benchmark. Hard P50/P95 numbers should come from telemetry before they are published.
Stack
A web-first voice stack for AI companions
| Layer | Provider / system | User-visible benefit | Privacy posture |
|---|---|---|---|
| Product surface | iTsChat web app | Mobile-first AI companion chat that starts on the web. | Policy links, age-gated entry, and protected-chat boundaries stay visible. |
| Speech-to-text | xAI STT, with provider-aware fallback support | Live beta testing has shown near-instant transcript updates, including low-signal 4G use. | The active STT provider is disclosed in the Privacy Policy and handled through server-owned configuration. |
| AI response | Grok / xAI runtime path | Companion replies are generated for fictional AI characters instead of generic assistant output. | Conversation context is scoped to product operation, continuity, billing, safety, and support needs. |
| Voice output | ElevenLabs text-to-speech | Companion replies can become spoken audio while keeping text-first recovery available. | TTS requests use the companion reply text needed to provide voice output. |
| Protected voice bridge | Split web app and Node voice service | Realtime voice work stays isolated from the marketing app and protected chat shell. | Telemetry is designed for operational timing and troubleshooting, not transcript publication. |
Voice pipeline
From spoken turn to companion voice
Mic capture
SystemBrowser microphone capture and voice activity handling
OutputAudio frames for the active turn
BoundaryUser action controls when voice capture starts or stops.
STT
SystemxAI speech-to-text path in the protected voice pipeline
OutputLive transcript updates and final turn text
BoundaryRaw voice is used to transcribe the request and run the service.
Runtime
SystemCompanion runtime and selected fictional persona
OutputText reply shaped for the active companion
BoundaryNo real person, escort, therapist, or professional advisor is represented.
TTS
SystemElevenLabs voice generation where voice is enabled
OutputAudio for the companion reply
BoundaryText remains available if audio is unavailable or paused.
Playback
SystemBrowser audio playback with user controls
OutputSpoken companion response
BoundaryThe user can pause playback or start a new mic turn.
Why it matters
Not a generic chatbot demo
Voice is part of the product shape
The system is designed around natural turn-taking: capture speech, settle the transcript, generate a fictional companion reply, and play voice without pretending audio is the only path.
Fast STT keeps the moment alive
The xAI STT path has felt close to immediate in live beta use, including low-signal 4G testing, so the user can see their spoken words become chat text without waiting through a cold workflow.
Policy links stay close
Privacy, Terms, and Billing Policy links remain public and crawlable so users and reviewers can inspect provider posture, adult-access rules, and billing boundaries.
FAQ
AI voice companion questions
What AI voice chat technology does iTsChat use?
iTsChat uses a web-first voice pipeline with browser microphone capture, provider-aware speech-to-text, AI companion response generation, ElevenLabs text-to-speech, and browser playback.
Does iTsChat support low-latency speech-to-text?
Yes. In live beta testing, the xAI speech-to-text path has shown near-instant transcript updates during natural voice chat, including low-signal 4G testing. iTsChat does not publish hard millisecond benchmarks unless they are backed by telemetry.
What makes iTsChat different from generic AI chatbots?
iTsChat is built around fictional AI companions, protected private chat, voice turn-taking, policy-aware adult access, billing readiness, and continuity seams instead of a generic assistant prompt box.
How does iTsChat handle voice privacy?
Voice input is used to convert speech into text and operate the voice-chat feature. Provider-specific STT language appears in the Privacy Policy, and protected chat surfaces are treated as sensitive product areas.
Read the operating boundaries
iTsChat companions are fictional AI-generated experiences, not real people or professional advisors. Before you continue, review the policy pages that describe privacy, adult access, billing, and service limits.