Apppublic

Abdul-Rehman-99/Browser-Voice_Agent-Backend

sourceHugging Faceupdated 3mo agoView on Hugging Face

1likes

App README

Browser Voice Agent Backend

Real-time voice AI backend with STT (Whisper/Groq), agent (Llama 3.3 / OpenAI Agents SDK), and TTS (Edge-TTS).

wss://<host>/ws/voice

Text frames (JSON control messages):

Direction	Type	Description
Client → Server	`start_session`	Begin conversation
Server → Client	`session_started`	Session confirmed
Client → Server	`audio_end` + `mime_type`	Mark audio complete
Server → Client	`processing`	Agent is thinking
Server → Client	`response_start`	TTS audio begins
Server → Client	`response_end`	TTS audio ends
Server → Client	`heartbeat`	Keepalive (respond with `pong`)
Server → Client	`turn_complete`	Full turn transcript + response

Binary frames: Raw audio chunks (bidirectional).

Variable	Default	Description
`GROQ_API_KEY`	—	Groq API key (required)
`GROQ_BASE_URL`	`https://api.groq.com/openai/v1`	Groq API base URL
`WS_HOST`	`0.0.0.0`	Bind address
`WS_PORT`	`8000`	Listen port
`AGENT_MODEL`	`llama-3.3-70b-versatile`	Groq model for agent
`STT_MODEL`	`whisper-large-v3-turbo`	Groq model for STT
`STT_LANGUAGE`	(auto)	Force STT language (`en`/`ur`/empty)
`TTS_VOICE`	`ur-PK-UzmaNeural`	Edge-TTS voice
`MAX_TURNS`	`15`	Conversation history turns

bash

uv sync
uv run uvicorn app.main:app --reload

Build and run with Docker:

bash

docker build -t browser-voice-agent-backend .
docker run -p 7860:7860 -e GROQ_API_KEY=your_key browser-voice-agent-backend