Question

nick_b_2016 on Mon, 14 Mar 2016 06:07:55


I am trying to stream data directly from the mic to the speech endpoint. However i continue to get timeouts after 14 seconds.

What exactly does "Your application must endpoint the audio to determine start and end of speech" this mean. I suspect i am not "endpointing" the raw audio data.

- Thanks.

Replies

cthrash99 on Mon, 14 Mar 2016 14:58:26


The 14-second limitation is there for the REST API, since the service is not designed to have a long-running streaming response to the client.  The error is saying that the onus is on the client to keep the audio shorter.  You can implement a speech detector, or have a push-to-talk UI to achieve this.

nick_b_2016 on Mon, 14 Mar 2016 15:55:46


Thanks for your reply!

I do understand that limitation, my UI is push to talk and my audio is well under 14 seconds. I am just trying to understand what "Your application must endpoint the audio to determine start and end of speech" means. I am getting hung up on "must endpoint". Is that simply saying your audio must stop... or do i need to insert something at the end of the audio stream that invariably says "DONE". I ask because I can send a pre-recorded file, but i cannot seem to correctly stream from the mic.

- Thanks!!!

cthrash99 on Mon, 14 Mar 2016 17:33:00


This is a case of unfortunate overload of the term 'endpoint.'  You do not need to have a stop word or the like.  You are, however, required to send audio that is cut at a proper audio boundary.  That's all that means.

It's difficult for me to know why the microphone audio isn't working for you.  But the error suggests that you're sending up more audio than you think.

nick_b_2016 on Mon, 14 Mar 2016 18:07:05


thx!