. Author manuscript; available in PMC: 2024 Mar 28.

Published in final edited form as: Proc ACM Interact Mob Wearable Ubiquitous Technol. 2023 Mar 28;7(1):12. doi: 10.1145/3580798

Table 11:

Performance comparison between manual transcripts and the ASR output. Weighted Average F₁-scores for activity and activity-stage recognition are reported. The number of words used in each case is shown in the first row.

	Entire vocabulary (4717)	Keyword-based (729)	Keyword-based (729) with data augmentation	ClinicalBERT+HuBERT with data augmentation
Utterance-level segmented & manual transcripts	69.3/75.1	69.6/76.8	72.0/77.4	68.1/74.8
Utterance-level segmented & ASR	58.7/61.2	60.0/62.7	60.9/65.2	54.1/59.3
Time window sliding & ASR	53.9/58.1	55.4/59.9	57.4/61.6	50.9/55.2