This tool visualizes word sequence alignments to facilitate spotting of errors in conversational ASR transcription
The example displayed on the poster presented at ICASSP 2024 Show&Tell.
A high-overlap (OV40) example from the LibriCSS dataset.
A high-overlap (OV40) example from the LibriCSS dataset, separated with a Continuous Speech Separation (CSS) pipeline and evaluated with the speaker agnostic tcORC-WER.
Error analysis on a DiPCo session processed by Paderborn University's submission to the 7th CHiME challenge.
Two synchronized visualizations to compare different systems. Here: NTT's and Paderborn University's submissions to the 7th CHiME challenge (DiPCo dataset).
Comparison of the tcp alignment (with time constraint) with the cp alignment (without time constraint).
The JupyterLite notebook that generates the visualization in the browser. Run it with you own data, no installation required!
Browse results across different sessions.
The poster presented at the ICASSP 2024 Show & Tell Demo session as PDF