Found Original Source of Error

[smmurphy]

Hi @MarkEdmondson1234 ,

I found the original source of the error which was apparently due to a missing confidence score on one line of the transcript. So, to address, I updated parse_async with the same timings and transcripts mappings you had for parse_speech, but added ifelse(!is.null(.x$confidence), .x$confidence,NA) to handle the missing confidence score. Missing confidence scores are apparently a thing: https://groups.google.com/forum/#!topic/cloud-speech-discuss/LUg7Ky_Pq3s. Considering that I added the same code for transcript and updated parse_speech with the same changes. While I was in there I also cleaned up the alert messages on parse_async. With this and the last pull request parse-async is called instead of parse_speech when the type includes LongRunningRecognize; parse-async should now have correct alerts and parse the result in the absence of confidence scores. I have not tested parse_speech, but the modifications there are minimal.

Note: parse_speech and parse_async are now quite similar. I'm not sure of the original intention of the separation and whether both are truly needed. Prior to the last pull request it seems that parse_speech was being called for asynchronous transcriptions and parse_async was unused.

[MarkEdmondson1234]

How annoying that the API is flakey like that, a bug I'd say but thanks to your work at least we can work around it. Thanks for getting to the root of the issue!

[smmurphy]

Great! My pleasure.