Chunk frequency

[Dmitriy Komarov]

Hello Arsen!

Lately we began to get an error "Chunks have to be sent with at least one per 5 seconds frequency" after wich session get drop. Yandex said that we need to send chunk at least one per 5 second with 5 second media.

As I understand there is no specific settings for chunk frequency in unimrcp.

2020-01-20 19:52:55:390959 [INFO]   Send MRCPv2 Data 10.101.132.227:1544 <-> 10.101.132.216:38696 [139 bytes]
MRCP/2.0 139 RECOGNITION-COMPLETE 13 COMPLETE
Channel-Identifier: 8b5180abefbc4540@speechrecog
Completion-Cause: 006 recognizer-error


2020-01-20 19:52:55:391043 [WARN]   gRPC Status: Chunks have to be sent with at least one per 5 seconds frequency <8b5180abefbc4540@yandexsr>

best regards, Komarov Dmitriy.

[San'cheZ]

Greetings, Dmitry.
I think these are Yandex restrictions. UniMRCP does not limit the number of requests. Except in cases of limitation of license.

вторник, 21 января 2020 г., 9:46:48 UTC+3 пользователь Dmitriy Komarov написал:

[Dmitriy Komarov]

Hello San'chez!

I think so - Yandex said that we need to send chunk at least once in 5 sec with 5 sec audio or they will drop session so we need to solve this on our side. In some situation - when we turn on possibly interrupting tts and after long tts message we try to answer something - session droping and we get error about chunk. There is one strange thing - I setup vad mode to 0 or 1 and in sessions logs I always see that it works in vad mode 2, why can this be?

Thank you for responding!

вторник, 21 января 2020 г., 22:52:52 UTC+3 пользователь San'cheZ написал:

[San'cheZ]

I think we need to restart the service. Something like systemctl restart unimrcp.

среда, 22 января 2020 г., 12:04:45 UTC+3 пользователь Dmitriy Komarov написал:

[Dmitriy Komarov]

Did that many times and of course restart whole server - it didn't help. Try to change to all mode - 0,1,3 but it still 2 in logs.

[Dmitriy Komarov]

Hi colleagues!
Can someone help with this problem?
It happens only when bargain on.

[Arsen Chaloyan]

Hi Dmitriy,

The problem description reminds me an issue investigated with Yandex and ultimately fixed by Yandex back in May/June 2019. You may refer to Jira UNI-50 while communicating this issue with the Yandex team. They should be able to identify the problem unless this is a related but not the exact same issue.

In fact, it would help if you could provide complete logs. So, we could see where the error " Chunks have to be sent with at least one per 5 seconds frequency" is originated from.

--
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/5d9803de-52d7-4551-9cd0-ebe9beac2fec%40googlegroups.com.

--

Arsen Chaloyan
Author of UniMRCP
http://www.unimrcp.org

[Dmitriy Komarov]

Hi Arsen,

Add full log of this session in attachment.

вторник, 28 января 2020 г., 5:56:38 UTC+3 пользователь Arsen Chaloyan написал:

Hi Dmitriy,

The problem description reminds me an issue investigated with Yandex and ultimately fixed by Yandex back in May/June 2019. You may refer to Jira UNI-50 while communicating this issue with the Yandex team. They should be able to identify the problem unless this is a related but not the exact same issue.

In fact, it would help if you could provide complete logs. So, we could see where the error " Chunks have to be sent with at least one per 5 seconds frequency" is originated from.

On Fri, Jan 24, 2020 at 2:12 AM Dmitriy Komarov <dmitriy.kom...@gmail.com> wrote:

Hi colleagues!
Can someone help with this problem?
It happens only when bargain on.

--
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 uni...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/unimrcp/5d9803de-52d7-4551-9cd0-ebe9beac2fec%40googlegroups.com.

[Arsen Chaloyan]

Hi Dmitriy,

This is exactly what I expected. It would help, though, if the YSR logging could have been set to DEBUG. Anyway, the problem must be fully reproducible as the same error has recently been reported from different sources, I beleive.

From what I can see, the service returns an error on second (or consecutive) recognition request in the same gRPC stream. Here is what is happening in more details. After the first recognition request completes, the user application plays a relatively long prompt to the caller. At the same time, the second recognition request is initiated. The caller starts speaking only when the prompt is finished. All is OK, nothing is really extraordinary. However, whenever gRPC config and audio chunk are sent, the service comes back with this the misleading status "Chunks have to be sent with at least one per 5 seconds frequency" in response to first audio chunk in a new utterance. Strange..

Here is the main sequence of events leading to this error condition.

2020-01-20 19:52:40:915828 [INFO]   Create gRPC Stream <8b5180abefbc4540@yandexsr>

2020-01-20 19:52:55:380472 [INFO]   Speech Detector State Transition NO-INPUT -> IN-PROGRESS [14470 ms] <8b5180abefbc4540>

2020-01-20 19:52:55:380647 [INFO]   Received Response: status [0] end-of-utt [0] result-count [0] <8b5180abefbc4540@yandexsr>

I believe the problem shall be addressed by the service provider in the same or a similar way the problem was originally fixed for the first recognition request in a session (gRPC stream).

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/bfaacda7-b740-46f2-806f-9db59c5fad2a%40googlegroups.com.