jupyter-bioacoustic 0.8.5

JupyterLab bioacoustic audio review plugin

JupyterBioacoustic

A JupyterLab plugin for reviewing and annotating bioacoustic audio clips.

Documentation

• site: https://schmidtdse.github.io/jupyter_bioacoustic/
• docs: https://github.com/SchmidtDSE/jupyter_bioacoustic/wiki

Table of Contents

• Install
• Quick Start
• Features
• Documentation
• Usage
• BioacousticAnnotator Parameters
• Demo
• License

---

Install

pip install jupyter-bioacoustic

See the Development wiki for building from source.

Quick Start

from jupyter_bioacoustic import BioacousticAnnotator

BioacousticAnnotator(
    data='detections-test.csv',
    audio='test.flac',
    info_card_title='[[common_name]]',
    form_config='form-review.yaml',
    output='reviews.csv',
).open()

See the Quick Start guide for test files and more examples.

Features

• Player / Visualizer — browse audio clips with interactive spectrograms and other visualizations (linear, mel, log-frequency, or custom)
• Configurable Forms — YAML-driven annotation and review forms with selects, textboxes, checkboxes, and conditional sections
• Annotation Tools — time markers, start/end lines, bounding boxes, and multibox (multiple labeled regions per clip)
• Built-In & Custom Visualizations — built-in linear and mel spectrograms, log-frequency, bandpass and waveform visualizations. Easily integrate third-party libraries (OpenSoundscape, librosa, SciPy, ...) or write your own
• Zoom and Capture — keyboard/mouse zoom, drag-to-pan, zoom-to-selection box, and PNG export
• Flexible data sources — CSV, Parquet, SQL (DuckDB), API endpoints, S3 byte-range reads
• Syncing — sync local annotation files to remote storage such as S3 and GCS

Documentation

Full documentation is on the wiki:

• Quick Start — Installation and first usage
• Configuration — All parameters, config files, capture, S3, kwargs
• Configurable Forms — YAML form layout reference
• Annotation Tools — Spectrogram interaction tools
• Data Schema — Input and output formats
• API Reference — BioacousticAnnotator class, properties, methods
• Audio IO — jupyter_bioacoustic.audio module reference
• Demo — Running the demo notebooks
• Development — Project structure, build tasks, architecture

Usage

The BioacousticAnnotator class has an extremely simple interface; having only two methods (.open(inline=True), .output(force=False)) and one property (.source).

from jupyter_bioacoustic import BioacousticAnnotator

# Create an instance
ja = BioacousticAnnotator(data='path_to_data.parquet', ...)

# Open the interface
ja.open()

# Get a dataframe with all the submitted data
# Note: this data is lazy loaded. this will read from
#       file each time you submit.
#       however between submissions it will be cached.
result_df = ja.output()
result_df = ja.output(force=True)  # force re-read from disk

# Dataframe access to the source data (here 'path_to_data.parquet')
ja.source

The parameters for BioacousticAnnotator are listed below. There is one special parameter config that can be used instead of providing the parameter values directly in the notebook. This is a great feature for reproduciblity, organization and avoiding bloated notebooks.

Consider the example above:

BioacousticAnnotator(
    data='detections-test.csv',
    audio='test.flac',
    info_card_title='[[common_name]]',
    form_config='form-review.yaml',
    output='reviews.csv',
)

This can instead be produced this way

BioacousticAnnotator(
    data='detections-test.csv',
    config='config/review-configuration.yaml',
).open()

# config/review-configuration.yaml
audio: 'test.flac'
info_card_title: '[[common_name]]'
form_config: 'form-review.yaml'
output: 'reviews.csv'

For this simple example, this might not seem helpful. However for more advanced configurations this is quite useful. Moreover, in the example above the review-form has a configuration file form-review.yaml. If using config the form can be included directly.

See Configuration for full details. Here is an advanced example:

# BioacousticAnnotator Args
audio: "audio_path"    # column name — auto-detected (no slashes or dots)
display_columns: ["common_name", "confidence", "start_time", "county", "audio_path"]
info_card_title: '[[common_name]]'
info_card_text: '[[confidence]] | [[county]] | [[start_time]]'
capture: 'Save Spectrogram'
capture_dir: 'spectrograms'


# Form
form_config:
    title:
      value: 'REVIEW CLIP'
      progress_tracker: true
    select:
      label: Is Valid
      column: is_valid
      required: true
      items:
        - label: 'yes'
          value: 'yes'
        - label: 'no'
          value: 'no'
          form: correction_form
    textbox:
      label: notes
      column: notes
    annotation:
      start_time:
        label: Start
        column: start_time
        source_value: start_time
      end_time:
        label: End
        column: end_time
        source_value: end_time
      tools: start_end_time_select
    correction_form:
      - select:
          label: verified name
          column: verified_common_name
          required: true
          items:
            path: data/categories.csv
            value: common_name
      - select:
          label: verif. confidence
          column: verification_confidence
          items:
            - low
            - medium
            - high
    submission_buttons:
      line: true
      next:
        label: Skip
      submit:
        label: Verify

BioacousticAnnotator Parameters

Parameter	Type	Default	Description
data	DataFrame / str / dict	required*	Input data. String: file path, URL, api::url, or SQL (SELECT ...). Dict: {path|url|uri|api|sql, secrets, columns}.
data_path	str	None	Explicit file path for data (overrides data source).
data_url	str	None	Explicit URL for data (overrides data source).
data_sql	str	None	Explicit SQL query for data (overrides data source).
data_api	str	None	Explicit API endpoint for data (overrides data source).
data_secrets	dict, list, or false	None	Auth for data loading. {key, value} pairs. Value: env:VAR, dialog, or literal. Set to false to opt out of global secrets fallback.
display_columns	list	[]	Columns for the clip table.
audio	str or dict	required*	Audio source. String: local path, URL/URI, or column name (auto-detected). Dict: {path|url|uri|column|sql|api|src, prefix, suffix, fallback, secrets, property, response_index}.
audio_src	str	None	Audio source string (auto-detected as path, URL, or column name). Same as passing a bare string to audio.
audio_path	str	None	Explicit local file path for audio (overrides audio source).
audio_url	str	None	Explicit URL for audio (overrides audio source).
audio_uri	str	None	Alias for audio_url.
audio_column	str	None	Explicit column name for per-row audio (overrides audio source).
audio_prefix	str	''	Prefix joined with / to audio paths.
audio_suffix	str	''	Suffix joined with / to audio paths.
audio_fallback	str	''	Fallback when audio is a column and the row value is empty.
audio_secrets	dict, list, or false	None	Auth for audio loading (same format as data_secrets). Set to false to opt out of global secrets fallback.
audio_sql	str	None	SQL query to resolve audio path. Requires audio_property.
audio_api	str	None	API URL to resolve audio path. Requires audio_property.
audio_property	str	None	Field/column to extract from SQL/API response as the audio path.
audio_response_index	int	1	1-based row index for SQL/API response (1 = first row).
secrets	dict or list	None	Global auth — fallback for data_secrets, audio_secrets, and output_secrets. Each can opt out by setting to false.
output	str or dict	''	Output file path or sync config dict. String: local path. Dict: {path, uri/url, sync_button, recursive, secrets}. See Output & Sync.
output_path	str	None	Explicit local output file path (overrides output string or output.path dict key).
output_url	str	None	Remote sync destination (overrides output.uri/output.url dict key).
output_uri	str	None	Alias for output_url.
output_sync_button	bool / str	None	Show a sync button (true = "Sync", string = custom label).
output_recursive	bool	None	Passed to io.write() for uploading directories.
output_secrets	dict, list, or false	None	Auth for sync uploads (same format as data_secrets). Set to false to opt out of global secrets fallback.
project	str	None	Project file path. No other config params allowed when set.
project_name	str	None	Widget header title. Auto-derived from project filename if not set.
form_config	dict / str	None	Form layout — YAML file, dict, or None for no form.
info_card_title	str	''	Info card title — supports [[column_name]] templates. Also used for capture filenames.
info_card_text	str	''	Info card text — supports [[column_name]] templates.
duplicate_entries	bool	False	Allow multiple submissions per row
default_buffer	int / float	3	Default buffer time in seconds around each clip
capture	bool / str	True	Capture button (False to hide, string for custom label)
capture_dir	str	''	Directory prefix for captures
capture_height	int	None	Capture image height in pixels. Defaults to player_height if not set.
spectrogram_resolution	int / list	[1000, 2000, 4000]	Spectrogram width in pixels. List for a dropdown selector, single value for fixed. Prefix an item with selected:: to set the default (e.g. [1000, 'selected::2000', 4000]).
visualizations	list	['linear', 'mel']	Visualization types for the dropdown. Built-in strings ('linear', 'mel', 'log_frequency', 'bandpass', 'waveform') or custom callables. See Custom Visualizations.
partial_download	bool	True	Use byte-range downloads for remote audio (requires ffmpeg/pydub). Set to False to always download and cache the full file.
width	str	'100%'	Inline widget width.
height	int	900	Inline widget height.
config	str	None	Path to YAML/JSON config file
**kwargs			Fixed columns in every output row

* data is not required if data_path, data_url, data_sql, or data_api is provided. audio is not required if audio_src, audio_path, audio_url, audio_uri, audio_column, audio_sql, or audio_api is provided.

Demo

Example notebooks are included in the demo/ directory. They require additional dependencies (ipyleaflet, shapely, seaborn, requests).

1. Install with demo dependencies

With pixi:

cd demo
pixi run -e demo jba lab

This launches JupyterLab inside the demo pixi environment with the data-rate limit pre-configured.

With pip:

pip install -e ".[demo]"
jupyter lab --ServerApp.iopub_data_rate_limit=1e10

2. Download audio files (one-time)

Audio files are not included in this repository (they are large FLAC files, ~50-100 MB each).

These are large files. It will likely take multiple minutes per file to download. For demo purposes, you can replace them with any FLAC audio file — the spectrograms will look different but the plugin works the same way.

With AWS CLI (faster):

cd demo
mkdir -p audio
aws s3 cp s3://dse-soundhub/public/audio/dev/20230522_200000.flac audio/test-default.flac --no-sign-request &
aws s3 cp s3://dse-soundhub/public/audio/dev/20230524_200000.flac audio/test1.flac --no-sign-request &
aws s3 cp s3://dse-soundhub/public/audio/dev/20230525_200000.flac audio/test2.flac --no-sign-request &
aws s3 cp s3://dse-soundhub/public/audio/dev/20230526_000000.flac audio/test3.flac --no-sign-request &
wait

With curl:

cd demo
mkdir -p audio
curl -o audio/test-default.flac https://dse-soundhub.s3.us-west-2.amazonaws.com/public/audio/dev/20230522_200000.flac
curl -o audio/test1.flac https://dse-soundhub.s3.us-west-2.amazonaws.com/public/audio/dev/20230524_200000.flac
curl -o audio/test2.flac https://dse-soundhub.s3.us-west-2.amazonaws.com/public/audio/dev/20230525_200000.flac
curl -o audio/test3.flac https://dse-soundhub.s3.us-west-2.amazonaws.com/public/audio/dev/20230526_000000.flac

3. Open a demo notebook

Open simple-example.ipynb from the JupyterLab file browser. The notebook demonstrates both review and annotation workflows.

License

BSD 3-Clause