How audio-editing-challenge/audio-editing-challenge-baseline works
This repository provides two MMAE audio-editing baselines: a single AuK model track and an LLM-routed agent track. Both take benchmark samples through a runner and produce edited audio and evaluator manifests. In the agent track, the router chooses and executes sequential DSP, separation, or generative-edit tools; the latter two use separate HTTP services. The graph includes documented capabilities whose implementation excerpts were not provided; those links are identified as documented rather than inferred from unsampled code.
Main entry points: router.py LLM router · run_agent_mmae.py Agent runner · run_single_mmae.py Single runner · dsp.py DSP tools
Components 17
Baseline tracks
- Single runner
run_single_mmae.py - Agent runner
run_agent_mmae.py
Agent orchestration
- Agent config
src/agent/config.py - LLM router
src/agent/router.py
Audio editing
- DSP tools
src/agent/tools/dsp.py - Audio utilities
src/agent/audio_utils.py
Model services
- SAM-Audio client
src/agent/clients/sam_audio_client.py - SAM-Audio service
src/services/sam_audio_server.py - AuK client
src/agent/clients/auk_client.py - AuK service
src/services/auk_server.py
Benchmark I/O
- MMAE manifests
src/mmae_io.py - Edited audio
Other components
- MMAE inputs
- AuK base model
- OpenAI-compatible LLM
- SAM-Audio model
- AuK generative model
Connections 19
- MMAE inputs → Single runner: provides samples
README.md - MMAE inputs → Agent runner: provides samples
README.md - Single runner → AuK base model: edits audio
README.md - Single runner → Edited audio: writes audio
README.md - Agent runner → Agent config: builds config
run_agent_mmae.py - Agent runner → LLM router: routes samples
run_agent_mmae.py - Agent runner → MMAE manifests: builds manifests
run_agent_mmae.py - Single runner → MMAE manifests: builds manifests
run_single_mmae.py - Agent runner → Edited audio: stores output
run_agent_mmae.py - LLM router → Agent config: reads settings
src/agent/router.py - LLM router → OpenAI-compatible LLM: requests tool calls
src/agent/router.py - LLM router → DSP tools: executes DSP
src/agent/router.py - DSP tools → Audio utilities: loads and saves
src/agent/tools/dsp.py - LLM router → SAM-Audio client: dispatches separation
src/agent/router.py - SAM-Audio client → SAM-Audio service: posts audio
src/agent/clients/sam_audio_client.py - SAM-Audio service → SAM-Audio model: runs separation
README.md - LLM router → AuK client: dispatches edit
src/agent/router.py - AuK client → AuK service: posts audio
src/agent/clients/auk_client.py - AuK service → AuK generative model: runs generation
README.md