WWW
As a researcher wanting to run experiments with TORGO, we need to design a leave one phrase out strategy for error-correction. TORGO has a high level of prompt overlap. To this end analyze:
Sentence level prompts
Word level prompts
Design a leave one phrase out training/evaluation scenario, with speakers in a leave-one-speaker out training/eval scenario for ASR
AC
A script or atleast an outline of a script to partition the dataset based on a valid train/test split that has been decided
WWW As a researcher wanting to run experiments with TORGO, we need to design a leave one phrase out strategy for error-correction. TORGO has a high level of prompt overlap. To this end analyze:
Design a leave one phrase out training/evaluation scenario, with speakers in a leave-one-speaker out training/eval scenario for ASR
AC A script or atleast an outline of a script to partition the dataset based on a valid train/test split that has been decided