view README @ 5:b523456082ca tip

Update path to dataset and reflect modified chunk naming convention.

Summary
-------------------
This software relates to results reported in Foster et al. (2015). Scripts are included for analysing annotator agreement and for obtaining baseline GMM results, using the CHiME-Home dataset. Finally, the scripts are provided for obtaining the annotations.

NB: The CHiME-Home dataset is not included. The dataset (recordings & annotations) is available at TODO.

The scripts were developed and tested on Linux environments, using Python 2.7.6. Portability to other environments is not expected.

Overview of directory structure and scripts
-------------------
annotation_scripts/
  Contains scripts used to obtain annotations
  annotationkit_create_annotation_protocol.py
    Prepares annotation protocol CSV from list of audio files
  annotationkit_create_annotation_protocol_wrapper.sh
    Wrapper shell script for annotationkit_create_annotation_protocol.py
  annotationkit_play_chunks.py
    Play back list of audio chunks, based on specified annotation protocol CSV and read annotation strings from terminal
  warmup_phase/
    Contains preliminary scripts used to create protocol files for annotator warm-up phases (cf. Foster et al., 2015)

analysis_of_annotations/evaluate_annotations.py
  Code relevant to analysis of CHiME-Home annotations (not audio)

gmm_baseline_experiments/
  Contains scripts for baseline GMM evaluation.
    run_experiments.py
	  Main script for baseline GMM evaluation

How-to
-------------------
Reproduce analysis of annotations reported in Foster et al. (2015) :-

  1) Modify evaluate_annotations.py, so that variable DatasetPath points to directory containing CHiME-Home dataset
  2) Execute evaluate_annotations.py
  The script writes textual analysis data to standard output; a figure pdf is written to analysis_of_annotations/figures subdirectory.

Reproduce baseline results reported in Foster et al. (2015) :-

  1) Modify run_experiments.py, so that Settings['paths']['chime_home']['basepath'] points to Directory containing CHiME-Home dataset; Settings['paths']['resultsdir'] and Settings['paths']['featuresdir'] should respectively point to desired directories for storing pickle files of data structures containing results and features
  2) Execute run_experiments.py
  The script writes a figure pdf to gmm_baseline_experiments/figures subdirectory.

Adapt annotation scripts for alternative purposes :-

  1) Modify annotationkit_create_annotation_protocol.py. The purpose of this script is to read a list of audio files from standard input and write CSV to standard output; as presently implemented, each input file is a audio segment from which a set of 4-second chunks are sampled (as defined in script header, and sampling without replacement). The list of chunks is written to an output CSV, and defines a `protocol' file for use with the annotation script annotationkit_play_chunks.py. Variables chunkDuration and sampleRate should match desired chunk duration and given audio sample rate.
  2) Modify annotationkit_play_chunks_wrapper.sh so that parameters chunkDuration, sampleRate are consistent with annotationkit_create_annotation_protocol.py. Further, modify parameters PermittedAnnotationCharacters, LoneAnnotationCharacters, LabelMap to reflect chosen label alphabet. NB: As implemented, using the protocol CSV, annotationkit_play_chunks.py seeks within each audio segment to locate each audio chunk, thus no explicit segmentation of input audio is required.



How to cite
-------------------
P. Foster, S. Sigtia, S. Krstulovic, J. Barker, M. D. Plumbley. "CHiME-Home: A Dataset for Sound Source Recognition in a Domestic Environment," in Proceedings of the 11th Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA), 2015.