Optimized Voice Agent

Deepgram Nova-3 STT (WebSocket) → Groq LLM (openai/gpt-oss-20b) → Deepgram Flux TTS (/v2/speak PCM)

idle
Streaming Pipeline Tracker
Direct streaming WebSocket pipeline with Deepgram Flux TTS PCM output
Status: Idle
1. VADWaiting

Endpointing (150ms)

2. STTWaiting

Deepgram Nova-3 WS

Real-time (0ms)

3. LLMWaiting

Groq gpt-oss-20b

4. TTSWaiting

Deepgram Flux (/v2/speak)

5. AudioWaiting

24kHz PCM Audio

User Transcript (Deepgram STT)

Waiting for user speech...

Agent Response (Groq LLM → Flux TTS)

Waiting for response generation...

Turn Latency Measurements
Streaming benchmarks (Deepgram Nova-3 STT + Groq TTFT + Flux TTFA)
Total Latency: p50: 0ms | p95: 0ms
Groq TTFT: p50: 0msFlux TTFA: p50: 0ms
TurnVAD (ms)STT (ms)Groq TTFTLLM TotalFlux TTFATotal (ms)
No turns recorded yet. Click 'Start Speaking' to test the optimized pipeline with Deepgram Flux TTS.
Streaming Pipeline Logs Console
Ready. Start an interaction to stream pipeline logs...