I was going through the GCP Data Engineer Cert sample questions and came across this one
You are working on an ML-based application that will transcribe conversations between manufacturing workers. These conversations are in English and between 30-40 sec long. Conversation recordings come from old enterprise radio sets that have a low sampling rate of 8000 Hz, but you have a large dataset of these recorded conversations with their transcriptions. You want to follow Google-recommended practices. How should you proceed with building your application?
A. Use Cloud Speech-to-Text API, and send requests in a synchronous mode.
B. Use Cloud Speech-to-Text API, and send requests in an asynchronous mode.
C. Use Cloud Speech-to-Text API, but resample your captured recordings to a rate of 16000 Hz.
D. Train your own speech recognition model because you have an uncommon use case and you have a labeled dataset.
I have chosen but looks like the answer is A. The reason for me to choose B is that the data is available in bulk as they have mentioned they have already recordings in bulk and bulk data should be transferred asynchronously.
Can someone please clarify?
1 Answers
