Resp-Agent

A multi-agent framework for respiratory sound diagnosis and generation using deep learning.

Installation

1. Install PyTorch with CUDA Support

pip install torch==2.8.0 torchaudio==2.8.0 --index-url https://download.pytorch.org/whl/cu128

2. Install resp-agent

pip install resp-agent -i https://pypi.org/simple/

This installs all dependencies for both inference and training (including deepspeed, wandb, matplotlib, etc.).

Quick Start

Python API

from resp_agent import BEATs, BEATsConfig
from resp_agent.diagnoser import run_diagnoser
from resp_agent.generator import run_generator

# Diagnose respiratory sounds
result = run_diagnoser(
    audio_dir="./audio",
    output_dir="./output",
    metadata_csv="./metadata.csv",
    config_path="./Diagnoser/config.yaml"
)

# Generate respiratory sounds
audio = run_generator(
    ref_audio="./reference.wav",
    disease="Asthma",
    out_dir="./generated",
    config_path="./Generator/config.yaml"
)

Command Line Interface

# Run diagnosis
resp-agent diagnose --audio_dir ./audio --output_dir ./output --metadata_csv ./metadata.csv --config ./Diagnoser/config.yaml

# Run generation
resp-agent generate --ref_audio ./ref.wav --disease Asthma --out_dir ./output --config ./Generator/config.yaml

# Start interactive chat agent
resp-agent chat --lang zh  # Chinese
resp-agent chat --lang en  # English

Requirements

• Python >= 3.8
• PyTorch >= 2.0.0 (recommended: 2.8.0 with CUDA 12.8)
• CUDA-capable GPU (recommended for training and inference)

Environment Variables

Set your DeepSeek API key for the Thinker agent:

export DEEPSEEK_API_KEY='your-api-key'

License

MIT License