RFW0115: Evaluate model results in more detail for STT_NS using speaker ID

[spsither]

RFW0115: Evaluate model results in more detail for STT_NS using speaker ID

Summary

Evaluate model performance on different categories within STT_NS using the segment ID and the speaker metadata Google Sheet

Key Concepts

STT_NS: Natural Speech Department in Speech To Text

Context

Use the speaker metadata Google sheet and get a more detailed breakdown of model performance within STT_NS.

Outputs

Detailed Report for the model performance in the STT_NS department. When there is more than one speaker in an audio file, we can use a common profile between the speakers and use that as the metadata.

Inputs

Benchmark dataset. Speaker metadata Google Sheet STT Model

Timeline

4 days

[kaldan007]

i think it is related to RFW0114. if we decide on the different sub category or filter within the departments, we can evaluate according to that.

[spsither]

Further ideas: Do speaker classification among the class in the Google Sheet. We can do clustering to get speaker-based classes.

https://arxiv.org/pdf/2109.15053.pdf