CF Voice Dialog

Diagnostics

Last turn i

Pipeline total (speech end → audio) i

—

Session averages i

Avg pipeline i —

VAD → first interim i —

Speech duration i —

STT final → LLM 1st i —

LLM 1st → TTS start i —

TTS start → audio i —

vLLM TTFT i —

TTS WS connect i —

TTS synthesis i —

Turns 0

Best / Worst i — / —

Segment min/max

VAD→interim—

STT→LLM—

LLM→TTS—

TTS→audio—

Bandwidth i

Audio uploaded i 0 KB

Audio downloaded i 0 KB

JSON overhead i 0 KB

Session startup i

WS connect i —

Mic init i —

Clock offset i —

Session

Messages sent i 0

Messages received i 0

Session # 0

Barge-in i —

Barge-ins i 0

AI model i —

Connection

Region i —

Voice i —

State i —

MIC i

Gain i 1.0× |

State — i

Press START to begin a session.

Connection

Worker URL i

JWT Secret i

User ID i

Session

Language i

Language controls both STT and TTS voice. German TTS: Deepgram Aura-2 DE supported. Russian TTS: not available from Deepgram — English voice used as fallback.

Barge-in i

OFF

AI behaviour i

System prompt i

Preview will appear here…

i Use @user_said as a placeholder for the user's latest sentence.

Temperature i

0.70 2

Max tokens i