Watson Speech to Text performance issue

2
votes

The recent issue with the Watson speech to text servers they are seeing is, one not being able to run audio files with duration greater then 10 minutes in a single instance and secondly there are times when TCP connection is lost with the Watson server when the speech to text is running for files which are less than 10 minutes.

They are basically using web sockets with the Watson server in python and wanted to understand what is the best way to run speech to text for audio files that have larger duration (for instance, we have had meetings going till 3 hours ). What would be the best way to configure the bottleneck for connection lost over web sockets as well.

1
ibm-cloudspeech-to-textibm-watson
Can you clarify who the "they" is that you are talking about? Is this a problem that you yourself are facing? Can you say more about the configuration that is being used, so that it is more clear why the connection is being lost? – ralphearle

1 Answers

4
votes

I've found that if you use WebSockets and do not ask for interim_results, the connection drops. A workaround until the issue gets fixed is to enable interim_results and ignore them until you get the final results. You can enable interim results by passing interim_results = true in the initialization message:

{
  "action" : "start",
  "content-type" : "...",
  "continuous" : true, 
  "interim_results" : true,
  "inactivity_timeout": -1, 
  ...
}