How microsoft/aec-challenge works
AEC Challenge combines audio datasets, baseline echo-cancellation inference, and objective AECMOS scoring. A typical researcher workflow is to use microphone and loudspeaker-loopback recordings as enhancement inputs, produce enhanced audio, then evaluate it against the original recordings. The README also describes an online subjective-test framework, but no implementation for it appears in the supplied tree. The graph covers documented capabilities and sampled baseline/scoring code; unsampled dataset internals and subjective-test wiring are not inferred.
Main entry points: icassp2022/enhance.py ICASSP 2022 enhancer · interspeech2021/enhance.py Interspeech 2021 enhancer · baseline Enhanced audio · aecmos.py AECMOS client
Components 14
Audio datasets
- Training recordings
datasets - Challenge test recordings
datasets - ICASSP 2021 samples
baseline/icassp2021
Echo cancellation
- ICASSP 2022 enhancer
baseline/icassp2022/enhance.py - Interspeech 2021 enhancer
baseline/interspeech2021/enhance.py - DEC inference
baseline/icassp2022/enhance.py - Enhanced audio
baseline
Evaluation
- AECMOS client
AECMOS/aecmos.py - Local AECMOS
AECMOS/AECMOS_local/aecmos.py - AECMOS ONNX models
AECMOS/AECMOS_local - Objective scoring service
- Subjective test framework
Signal processing
- Audio utilities
baseline/interspeech2021/audio_utils.py
Other components
- Researcher
Connections 14
- Researcher → Training recordings: uses data
README.md - Researcher → Challenge test recordings: uses data
README.md - Researcher → ICASSP 2022 enhancer: runs inference
baseline/icassp2022/enhance.py - Researcher → Interspeech 2021 enhancer: runs inference
baseline/interspeech2021/enhance.py - ICASSP 2022 enhancer → DEC inference: configures
baseline/icassp2022/enhance.py - Interspeech 2021 enhancer → Audio utilities: calls
baseline/interspeech2021/enhance.py - ICASSP 2022 enhancer → Enhanced audio: writes
baseline/icassp2022/enhance.py - Interspeech 2021 enhancer → Enhanced audio: writes
baseline/interspeech2021/enhance.py - Researcher → AECMOS client: scores audio
AECMOS/aecmos.py - AECMOS client → Objective scoring service: posts audio
AECMOS/aecmos.py - Enhanced audio → AECMOS client: evaluates
- Researcher → Local AECMOS: runs scoring
AECMOS/AECMOS_local/aecmos.py - Local AECMOS → AECMOS ONNX models: loads model
AECMOS/AECMOS_local/aecmos.py - Researcher → Subjective test framework: submits results
README.md