Supervisely

Overview

Import audio recordings with their segments, whole-recording tags and the project's spectrogram settings in the Supervisely format. A downloaded Audio project is already in this format.

All information about the Supervisely audio annotation format can be found here.

Format description

Supported audio formats: .wav, .flac, .mp3, .ogg, .m4a With annotations: yes Supported annotation format: .json Data structure: Information is provided below.

Input files structure

Both a directory and an archive are supported. Each dataset is a directory with audio and ann subdirectories; nested datasets are stored in a datasets subdirectory of their parent.

📦 project_name
├── 📄 meta.json
└── 📂 dataset_name
    ├── 📂 audio
    │   ├── 🎵 recording_01.wav
    │   └── 🎵 recording_02.flac
    ├── 📂 ann
    │   ├── 📄 recording_01.wav.json
    │   └── 📄 recording_02.flac.json
    └── 📂 datasets
        └── 📂 nested_dataset_name
            ├── 📂 audio
            └── 📂 ann

What is imported

• Recordings, unchanged.

• Segments and whole-recording tags, with their values, channels, meta and customData, matched to the destination project's tags by name. A tag that conflicts with an existing one is renamed, as for other project types. Ids from the source are replaced.

• Datasets, including nested ones.

• Spectrogram settings from meta.json, but only when the destination project has none yet and holds no recordings. A project that is already configured, or already has recordings, keeps its own settings, because changing them would change what its existing labels mean; the import log says when the imported settings differ.