I understand the server waits for more input data after user done speaking. But in my case a user says a lot of short sentences or words, some words are especially slow, such as "underscore".
I tested with the model vosk-model-en-us-0.22-lgraph(128M)
with no phrase_list set, it takes about 10~15 seconds to get the result of "underscore"
with a 200 word phrase_list, it takes about 1+ second.
I tried to send a chunk of 40k binary zeroes to server, I think maybe the server considers it as a five-seconds-silence, then it may terminates the waiting and return the result immediately.
It does work for the 200 phrase_list, with the zeroes, it waits less than 500ms, comapred to previously 1+ second.
But it doesn't work for the full phrase_list mode , it still takes 10+ seconds. I notice that when it is calculating the "underscore", one of my CPU core keeps at 80%
I have two questions:
Why this word "underscore" takes so long?
Is there a built-in way to stop waiting?
BTW, I made a tool that demostrates how to capture sound from microphone and send to VOSK server, it can also play back and save input to a .wav file for troubleshooting, written in Golang and cross platform.
In case anyone wants an interactive demo: https://github.com/aj3423/vosk-sound-test/
And I'm sure there is no sound quality problem in my case..
I understand the server waits for more input data after user done speaking. But in my case a user says a lot of short sentences or words, some words are especially slow, such as "underscore".
I tested with the model vosk-model-en-us-0.22-lgraph(128M)
phrase_list
set, it takes about 10~15 seconds to get the result of "underscore"phrase_list
, it takes about 1+ second.I tried to send a chunk of 40k binary zeroes to server, I think maybe the server considers it as a five-seconds-silence, then it may terminates the waiting and return the result immediately. It does work for the 200
phrase_list
, with the zeroes, it waits less than 500ms, comapred to previously 1+ second.But it doesn't work for the full
phrase_list
mode , it still takes 10+ seconds. I notice that when it is calculating the "underscore", one of my CPU core keeps at 80%I have two questions:
BTW, I made a tool that demostrates how to capture sound from microphone and send to VOSK server, it can also play back and save input to a .wav file for troubleshooting, written in Golang and cross platform. In case anyone wants an interactive demo: https://github.com/aj3423/vosk-sound-test/
And I'm sure there is no sound quality problem in my case..