Temporary no-input-timeout responses

Matthias B.

unread,

Dec 15, 2022, 11:58:55 AM12/15/22

to UniMRCP

We offer an automatic booking system for our customers based on an asterisk phone system, utilizing unimrcp as an interface to google speech recognition/synthesis. At times we have the problem that we receive a no-input-timeout message. As we record the corresponding calls, i can tell for sure that there is speech input from the caller, but no response from unimrcp except that message.

This error always occurs at time periods, not continuously. For example today we 5 hours error free and then 5 errors in 30 minutes.

Another observation is, if there are more recognitions running at the same time, it will not affect all recognitions. Only one of them produces that error.

After searching in this group we have modified the default values of the parameters parameters in order to improve the recognition. At the moment we have the following settings:

<speech-dtmf-input-detector
vad-mode="2"
speech-start-timeout="10"
speech-complete-timeout="1000"
speech-incomplete-timeout="15000"
noinput-timeout="15000"
input-timeout="30000"
dtmf-interdigit-timeout="5000"
dtmf-term-timeout="10000"
dtmf-term-char=""
speech-leading-silence="25"
speech-trailing-silence="100"
speech-output-period="50"
/>

<streaming-recognition
api="v1"
model="phone_call"
use-enhanced="true"
language="de-DE"
region="eu"
service-uri="eu-speech.googleapis.com"
grpc-log-redirection="true"
grpc-log-verbosity="DEBUG"
grpc-log-trace="http"
single-utterance="true"
interim-results="true"
start-of-input="service-originated"
max-alternatives="1"
alternatives-below-threshold="false"
sort-alternatives="false"
confidence-format="auto"
results-indent="2"
skip-empty-results="true"
skip-unsupported-grammars="true"
transcription-grammar="transcribe"
grammar-param-separator=";"
profanity-filter="false"
word-time-offsets="false"
word-confidence="false"
auto-punctuation="false"
spoken-punctuation="false"
spoken-emojis="false"
speaker-diarization="false"
min-speaker-count="2"
max-speaker-count="6"
stream-creation-timeout="0"
inter-result-timeout="0"
tag-format="default"
/>

Barge in is enabled, what improved the recognition a lot!

We have turned DEBUG on for the gsr plugin, hence we could see, that there is a constant grpc message stream to the google server, even in case of the error condition. In case of error there is merely no response from the google server.

Does anybody have any idea?

Best regards

Matthias Bremer

Vahagn Kocharyan

unread,

Dec 21, 2022, 6:08:35 PM12/21/22

to UniMRCP

Hi

please share your logs for addtional investigation

Arsen Chaloyan

unread,

Dec 22, 2022, 8:34:45 PM12/22/22

to uni...@googlegroups.com

Hi Mattias,

I'd first enable collection of utterances and listen to what is actually streamed to Google in those cases where there is no response. You can do that by enabling the following parameter in umsgsr.xml. Utterances will be stored in /opt/unimrcp/var directory by default

<utterance-manager
save-waveforms="true"

If you could share the logs with recorded utterances, that may allow to comment further... Otherwise, there are no inherit problems in your setup and everything should work flawlessly all the time.

--
You received this message because you are subscribed to the Google Groups "UniMRCP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unimrcp+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unimrcp/57b8109d-9b23-4633-8e4a-aeb611f11ccbn%40googlegroups.com.