Google Cloud Speech API speed up

[Andrey Malygin]

We are using google cloud speech for phone voice bot. 

To conduct a natural dialogue with a virtual operator the speed of speech recognition is one of the crucial parameter for us.

We send wav file with 2-3sec in duration and 33k in size on average. Speech Recognition by google takes about 2-5 sec. 

Can we reduce the response time up to 0.1-0.5 sec per request?

[Yannick (Cloud Platform Support)]

Hello Audrey, have you tried using Streaming Speech Recognition? This is the best way to get fast response times from the Speech API as it save the Speech API from having to listen to the audio after the fact.

[Guy Jerby]

Hi,

I have the same scenario, I changed it from file to stream, it reduced from 5 seconds to 3 seconds, but it is still too slow.

I am trying now to split the stream to 100-milliseconds chunks instead of 1 second chunks - it should help - but I didn't manage to run it with 100 millisecond chunks.

is anyone also have this case?

Thanks,

Guy

[Евгений Пашкелевич]

I use example from your link.  all this small application takes about 20 seconds to create client and respond me text.  I use local wav file. 1 channel 16000 rate. size 90 kb.   20 sekonds.  it's too much.   how can I get at least 3? I tried stream, suncro or asynchro commands... the same result. in average.  internet speed is 20mb/sek 

пятница, 27 апреля 2018 г., 16:08:13 UTC+3 пользователь Guy Jerby написал:

[Monica (Google Cloud Platform)]

There is a best practices documentation available [1]. Also have a look at what is suggested in this StackOverflow thread [2].

Can you further clarify your question: "how can I get at least 3?"

[1] https://cloud.google.com/speech-to-text/docs/best-practices 

[2] https://stackoverflow.com/questions/50364955/how-to-speed-up-google-cloud-speech