Watson Speech to Text should return timestamps accurate to milliseconds for transcription

Real-life scenario:

Researchers from Brandeis University, Boston University, Harvard, Boston College, and Northeastern University are investigating cognitive aging and biomarkers of dementia. Currently, they have been hand-scoring cognitive interviews which is arduous. To automate some of the manual work, Watson Text to Speech service is being used to run audio transcriptions through the service while getting back transcriptions with timestamps of when a word starts to be spoken and when it is ended in speech. The timestamps returned from the service return with 2 decimal places which is not enough precision that is required to compare to prior work (which uses 3 decimal places -- precision is up to milliseconds).

Problem statement:

The issue that researchers mentioned above are running into is comparing transcription timestamps generated by Speech to Text service to hand-scoring done for interviews prior to using Watson. The precision mismatch does not allow the researchers to use Watson effectively.

Current workaround:

No workaround is possible since there is a mismatch in precision level returned by Watson Speech-to-text service.

Proposed solution:

Having timestamps returned with 3 decimal places (with millisecond precision) would enable the researchers to automate the longitudinal research with Watson Speech-to-Text service.

Benefits/Value:

Watson Speech-to-Text service returning timestamps with milliseconds precision will enable customers around the world to get more accurate transcriptions and use the service more confidently in research globally. With precision, researchers will be more likely to use the service and cite it in publications.

Users impacted:

Every user using Watson Speech-to-Text around the world will get more precise data points for transcription without their current downstream applications breaking due to this change.