AudioPreprocessing - AMI corpus

This program processes the AMI corpus dataset of meetings in order to create segmented audio files for each speaker by referencing the annotations of the data.

Prerequisites

• Download the corpus from here
• Download the public manual annotations here
• Set the path parameters in the config files as follows,

  "directories": {
    "Audio": {
      "type": "<DOWNLOADED_AUDIO_TYPE>",
      "path": "<AMI_CORPUS_PATH>"
    },
    "Segments": {
      "path": "<PATH_OF_SEGMENTS_FOLDER_IN_PUBLIC_MANUAL>"
    },
    "Words": {
      "path": "<PATH_OF_WORDS_FOLDER_IN_PUBLIC_MANUAL>"
    },
    "Meetings": {
      "path": "<PATH_OF_MEETINGS_XML_IN_PUBLIC_MANUAL>"
    },
    "Participants": {
      "path": "<PATH_OF_PARTICIPANTS_XML_IN_PUBLIC_MANUAL>"
    }
  }
  "OutputFolder": {
  "path": "<PATH_OF_THE_OUTPUT_FOLDER>"
  },
  "CombinedOutputFolder": {
  "path": "<PATH_OF_THE_COMBINED_FILES_FOLDER>"
  },
  "SpeakerOutputFolder": {
  "path": "<PATH_FOR_THE_SPEAKER_FOLDER>"
  }

Preprocessing structure

How to use

• run Segment_audio.py
• To combine the audio files according to meeting participants, run Analyse_speaker.py
• To remove shorter audio segments and rename the files according to speaker, run Reformat_Files.py
• To combine the segmented audio files of each meeting according to each speaker, run Combine_folders.py