VoicePlus

Realtime voice assistant prototype for phone calls and spoken conversations. VoicePlus captures two audio streams, sends low-latency transcription to Speechmatics, displays live S1/S2 transcript lines, and can generate short response suggestions through local Ollama models or optional OpenAI helpers.

Features

• Dual-stream call transcription for speaker/customer sides.
• Speechmatics realtime WebSocket transcription.
• Terminal UI with live partials, final transcript history, pause state, and colored output.
• Echo suppression heuristics between S1 and S2 streams.
• Optional translation pipeline.
• Local LLM reply suggestions through reply_engine.py and Ollama.
• Optional OpenAI helper calls for translating a Russian goal into English and generating useful phrases.
• Per-call logs written to call_logs/.

Files

• main.py is the current entry point.
• reply_engine.py contains local LLM prompt routing and cleanup logic.
• vp.py and vp_2.py are earlier application variants kept for reference.
• test.py and test_2.py are local benchmark/scratch scripts for model latency testing.

Configuration

Create a local .env file from .env.example:

cp .env.example .env

Required for transcription:

SPEECHMATICS_API=replace-with-speechmatics-api-key
SPEECHMATICS_WSS=wss://eu2.rt.speechmatics.com/v2/

Optional AI helpers:

OPENAI_API=replace-with-openai-api-key
OLLAMA_HOST=http://127.0.0.1:11434
ENABLE_SUGGEST=1

API keys and call logs must stay local. Do not commit .env, logs, audio captures, or generated transcripts.

Development

Install dependencies:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Run:

python main.py

The app expects local audio routing/capture to be configured on the host. For local suggestion generation, run Ollama and pull the models referenced by reply_engine.py.

Status

This is an experimental realtime assistant. The public version keeps runtime configuration external and focuses on the transcription, UI, and response-suggestion workflow.