An integrated audio annotation pipeline to help researchers and developers annotate their speech data developed by TMH
- Breath and silence detection
- Speaker diarization
- Overlap detection (TODO)
- Laughter detection (TODO)
- Automatic Speech recognition (ASR) (TODO)
Speech Unlock offers pre-trained models as well as training capabilities for them
Another feature that Speech Unlock provides is the ability to generate and host ASR crowdsourcing tasks for Prolific, Amazon Mechanical Turk
The file format that is central for Speech Unlock and is
used throughout the pipelines is called a lab file and
has a .lab extension.
The lab file contains multiple files and each line contains one
temporal annotation segment (for example a continuous
time period within a file where a person laughed) and has the
following format: <start_sec> <end_sec> <text> where <start_sec> is the start of the segment
in seconds, <end_sec> is the end of the segment is seconds and
<text> is the label of the annotation segment.
To set up the pip dependencies install pip requirements on a clean environment:
pip install -r requirements.txt