Praditor

A DBSCAN-Based Automation for Speech Onset Detection and Voice Activity Detection

Download Praditor | English · 中文 | Our Paper

Features

Praditor is a speech onset detector that automatically finds boundaries between silence and sound. It supports two detection modes and generates output in .TextGrid format — ready for use in Praat.

Tip

Need automatic time & content annotation? Check out Praasper — a VAD-Enhanced Automatic Speech Annotation pipeline for Psycholinguistic Research.

• Onset/Offset Detection (Default Mode) — Detects the start and end of sound events using DBSCAN clustering and first-derivative thresholding. Outputs PointTier layers for onsets (blue) and offsets (green).
• Voice Activity Detection (VAD Mode) — Detects speech segments and outputs an IntervalTier with "sound" intervals. VAD mode uses fixed kernel parameters and processes audio in 15-second segments with intelligent boundary detection.
• Batch Processing — Run All processes every audio file in the current folder, generating .TextGrid and CSV summary files.
• Parameter Persistence — Three save modes (Default / Folder / File) with priority-based loading. VAD mode parameters are stored separately with _vad suffix.
• Parameter History — Navigate through up to 10 previous parameter sets with Backward / Forward buttons.
• CSV Export — Automatic per-folder CSV summary with timestamps for all processed audio files.
• Real-time Preview — Test button estimates onset/offset count without modifying .TextGrid output.

Try test_audio.wav and test_audio_mp3.mp3 on Praditor. test_audio_mp3.mp3 is from an online resource, while test_audio.wav and test_large_audio.wav are from our own experiments.

GUI Overview

Titlebar

Button	Action
File	Import audio files (.mp3, .wav, .ogg, .aac, .flac, .amr, .wma, .aiff)
? (Help)	Open documentation
Run	Run detection on the current file
Run All	Batch process all audio files in the current folder
Test	Preview how many onsets/offsets would be detected with current parameters (does not modify .TextGrid)
Stop	Cancel running detection
Trash	Clear displayed annotations (does not modify .TextGrid)
Read	Reload annotations from .TextGrid file
Onset	Toggle onset annotations (blue)
Offset	Toggle offset annotations (green)
◀ / ▶	Navigate to previous / next audio file

Audio Viewer

Displays waveform with onset/offset markers and time labels. Supports zoom and scroll via mouse, keyboard, and touchpad.

Mouse & Keyboard

• Wheel ↑ / Wheel ↓ — Zoom amplitude
• Ctrl + Wheel ↑ / Wheel ↓ — Zoom timeline
• Shift + Wheel ↓ / Wheel ↑ — Scroll timeline
• F5 — Play / pause audio; any key to stop

Touchpad

• Vertical scroll — Zoom amplitude
• Horizontal scroll — Zoom timeline
• Horizontal swipe — Scroll timeline

Timeline zoom may not work on macOS touchpad. Use Command + I / Command + O instead.

Parameter Sliders

Nine parameters for onset (blue) and nine for offset (green), independently adjustable:

Parameter	Description
Threshold	Coefficient for actual threshold (baseline × coefficient)
NetActive	Accumulated net count of above-threshold frames required
Penalty	Penalty for below-threshold frames
RefLen	Length of reference segment for baseline calculation
KernelFrm%	Percentage of frames retained in the kernel
KernelSize	Kernel size in frames
EPS%	Neighborhood radius in DBSCAN clustering
LowPass	Lower cutoff frequency of bandpass filter
HighPass	Higher cutoff frequency of bandpass filter

In VAD mode, offset sliders are hidden and Penalty, RefLen, KernelFrm%, KernelSize are fixed to VAD-optimized defaults.

Toolbar

Button	Action
Default	Use application-level default parameters
Folder	Use folder-level parameters (params.txt / params_vad.txt in current folder)
File	Use file-specific parameters (same name as audio file, .txt extension)
Save	Save current parameters to selected mode(s)
Reset	Reload parameters from selected mode's saved file
Backward / Forward	Navigate parameter history (up to 10 sets per mode)
VAD	Toggle VAD mode
1/10	Current parameter index / total

Priority: File > Folder > Default. When multiple modes are selected, Save writes to all selected locations; Reset loads from the highest-priority location that exists.

Video Instruction

Fine-tuning Guidance

Basic understanding is enough. Understanding the algorithm is better.

• Basic knowledge: Go to the first section of Quick Fix.
• Advanced knowledge: Go to the second section of Quick Fix (Detailed Introduction).
• Expert knowledge: Go to Parameters.

Data and Materials

If you would like to download the datasets that were used in developing Praditor, please refer to our OSF storage.

Citation

If you use Praditor in your research, please cite the following paper:

Liu, Z., Yu, X., Hu, W.C. et al. Praditor: A DBSCAN-based automation for speech onset detection. Behav Res 57, 247 (2025). https://doi.org/10.3758/s13428-025-02776-2

Or download .ris from the paper's About this article page.

Acknowledgments

Shout out to these remarkable contributors!

• Thank YU Xinqi, Dr. MA Yunxiao, ZHANG Sifan for their work in validating the effectiveness of Praditor's algorithm.
• Thank HU Wing Chung for her work in packaging Praditor for macOS (arm64 and universal2).
• Thank Prof. ZHANG Haoyun (University of Macau) and Prof. WANG Ruiming (South China Normal University) for their guidance and support.

Also, the funding:

• This project was funded by the National Natural Science Foundation of China (32200845), the Science and Technology Development Fund, Macao S.A.R (FDCT, 0153/2022/A), and the Multi-Year Research Grant (MYRG2022-00148-ICI) from the University of Macau to Haoyun Zhang.

Contact

Praditor is written and maintained by Tony, Liu Zhengyuan from the Centre for Cognitive and Brain Sciences, University of Macau.

If you have questions about using Praditor, its algorithm details, or need custom scripts (audio export, Excel tables, etc.), feel free to contact me at zhengyuan.liu@connect.um.edu.mo or paradeluxe3726@gmail.com.

License

Praditor is dual-licensed under AGPL v3 + a commercial license (LICENSE):

• AGPL v3 (default): free, open source. Academic / personal / non-profit / small orgs can use it directly. Only requirement: if you offer Praditor as a network service, you must make the source available.
• Commercial License: if you cannot accept AGPL copyleft obligations (e.g. commercial products, SaaS, large org internal use), purchase a commercial license to waive AGPL terms.

See COMMERCIAL-LICENSE.md for details.