Whisper turns audio into text. Max about MB per file. Recording needs HTTPS and microphone permission. "Translate to English" works with any spoken language.
0:00or
Chars 0Words 0~Tokens 0
Orpheus speaks your text. Each request is limited to characters, so longer text is split and joined automatically (up to characters per click). Output is WAV. English supports style tags like [cheerful] or [whisper].
Arabic model: if you get an access error, open the Groq Console and accept the Canopy Labs model terms for your account first.
Click the mic, speak, click again. Flow: Whisper (your voice → text) → AI chat (api.php) → Orpheus (reply → voice). Replies are asked to be short, and only the first ~400 characters are spoken (full reply stays as text).
Click the mic to start
Prompt Guard detects prompt-injection / jailbreak attempts (score 0 to 1; 0.5 or more = likely attack). It works best on short text. Safeguard checks text against your own policy (abuse, scams, etc.). Use these to screen visitor messages before sending them to the AI.