Missing inputs - GDF/GSR plugins
58 views
Skip to first unread message
Alexander Demidov
Feb 28, 2023, 9:37:53 AM2/28/23
to UniMRCP
Hello!
I'm using UniMRCP + FreeSwitch (mod_unimrcp) with Google Speech Recognition plugin and Google Dialog Flow plugin.
UniMRCP v1.7.0
GSR Plugin v1.22.1
GDF Plugin v1.20.1
I faced with a missing inputs in GDF and GSR.
I'm receiving detect-speech event with empty body, despite the fact, that MRCP server logs contain the correct transcription of the user input.
It happens quite often, so I believe that it's not an extraordinary case.
In UniMRCP server logs I have next lines:
2023-02-27 15:01:41:951005 [INFO] Speech Detector State Transition NO-INPUT -> IN-PROGRESS [1170 ms] <ef0dedaf887245f8>
2023-02-27 15:01:41:952432 [INFO] Start Input Timer [12000 ms] <ef0dedaf887245f8>
2023-02-27 15:01:41:952650 [INFO] Enable Output Audio: encoding=1, sampling-rate=8000 <ef0dedaf887245f8@gdf>
2023-02-27 15:01:41:952657 [INFO] Init Streaming Request: encoding=1, sampling-rate=8000, language=pl-PL, single-utterance=0, word-info=0 <ef0dedaf887245f8@gdf>
2023-02-27 15:01:41:952661 [INFO] Set Model [default] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:077523 [INFO] Received Response: status [1] recog result [1] detect intent response [0] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:077557 [INFO] Recognition Result: transcript [imię] confidence [0.00] final [0] end-offset [1:320] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:077577 [INFO] Set Result Flag [1000 ms] <ef0dedaf887245f8>
2023-02-27 15:01:43:077612 [INFO] Process START-OF-INPUT Event <ef0dedaf887245f8@speechrecog> [4]
2023-02-27 15:01:43:077641 [INFO] Send MRCPv2 Data 10.168.8.111:1544 <-> 10.168.8.83:55484 [115 bytes]
MRCP/2.0 115 START-OF-INPUT 4 IN-PROGRESS
Channel-Identifier: ef0dedaf887245f8@speechrecog
Input-Type: speech
2023-02-27 15:01:43:206761 [INFO] Received Response: status [1] recog result [1] detect intent response [0] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:206808 [INFO] Recognition Result: transcript [Kim jest] confidence [0.00] final [0] end-offset [1:440] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:579055 [INFO] Received Response: status [1] recog result [1] detect intent response [0] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:579095 [INFO] Recognition Result: transcript [imię] confidence [0.00] final [0] end-offset [1:800] <ef0dedaf887245f8@gdf>2023-02-27 15:01:45:640933 [INFO] Speech Detector State Transition IN-PROGRESS -> COMPLETE [3690 ms] <ef0dedaf887245f8>
2023-02-27 15:01:45:640964 [INFO] Detector Stats: leading-silence=300 ms, input=2820 ms, trailing-silence=1000 ms <ef0dedaf887245f8>
2023-02-27 15:01:45:641204 [INFO] Input Complete [success] size=65920 bytes, dur=4120 ms <ef0dedaf887245f8@gdf>
As you can see, it contains transcript [imię] , but I'm not receiving that input in the detect-speech event (body is empty).
How it can be fixed?
I'm using UniMRCP + FreeSwitch (mod_unimrcp) with Google Speech Recognition plugin and Google Dialog Flow plugin.
UniMRCP v1.7.0
GSR Plugin v1.22.1
GDF Plugin v1.20.1
I faced with a missing inputs in GDF and GSR.
I'm receiving detect-speech event with empty body, despite the fact, that MRCP server logs contain the correct transcription of the user input.
It happens quite often, so I believe that it's not an extraordinary case.
In UniMRCP server logs I have next lines:
2023-02-27 15:01:41:951005 [INFO] Speech Detector State Transition NO-INPUT -> IN-PROGRESS [1170 ms] <ef0dedaf887245f8>
2023-02-27 15:01:41:952432 [INFO] Start Input Timer [12000 ms] <ef0dedaf887245f8>
2023-02-27 15:01:41:952650 [INFO] Enable Output Audio: encoding=1, sampling-rate=8000 <ef0dedaf887245f8@gdf>
2023-02-27 15:01:41:952657 [INFO] Init Streaming Request: encoding=1, sampling-rate=8000, language=pl-PL, single-utterance=0, word-info=0 <ef0dedaf887245f8@gdf>
2023-02-27 15:01:41:952661 [INFO] Set Model [default] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:077523 [INFO] Received Response: status [1] recog result [1] detect intent response [0] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:077557 [INFO] Recognition Result: transcript [imię] confidence [0.00] final [0] end-offset [1:320] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:077577 [INFO] Set Result Flag [1000 ms] <ef0dedaf887245f8>
2023-02-27 15:01:43:077612 [INFO] Process START-OF-INPUT Event <ef0dedaf887245f8@speechrecog> [4]
2023-02-27 15:01:43:077641 [INFO] Send MRCPv2 Data 10.168.8.111:1544 <-> 10.168.8.83:55484 [115 bytes]
MRCP/2.0 115 START-OF-INPUT 4 IN-PROGRESS
Channel-Identifier: ef0dedaf887245f8@speechrecog
Input-Type: speech
2023-02-27 15:01:43:206761 [INFO] Received Response: status [1] recog result [1] detect intent response [0] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:206808 [INFO] Recognition Result: transcript [Kim jest] confidence [0.00] final [0] end-offset [1:440] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:579055 [INFO] Received Response: status [1] recog result [1] detect intent response [0] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:579095 [INFO] Recognition Result: transcript [imię] confidence [0.00] final [0] end-offset [1:800] <ef0dedaf887245f8@gdf>2023-02-27 15:01:45:640933 [INFO] Speech Detector State Transition IN-PROGRESS -> COMPLETE [3690 ms] <ef0dedaf887245f8>
2023-02-27 15:01:45:640964 [INFO] Detector Stats: leading-silence=300 ms, input=2820 ms, trailing-silence=1000 ms <ef0dedaf887245f8>
2023-02-27 15:01:45:641204 [INFO] Input Complete [success] size=65920 bytes, dur=4120 ms <ef0dedaf887245f8@gdf>
As you can see, it contains transcript [imię] , but I'm not receiving that input in the detect-speech event (body is empty).
How it can be fixed?
Arsen Chaloyan
Mar 10, 2023, 3:59:19 PM3/10/23
to uni...@googlegroups.com
Hello Alexander,
Here are all the interim results received from Google extracted from your logs.
2023-02-27 15:01:43:077557 [INFO] Recognition Result: transcript [imię] confidence [0.00] final [0] end-offset [1:320] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:206808 [INFO] Recognition Result: transcript [Kim jest] confidence [0.00] final [0] end-offset [1:440] <ef0dedaf887245f8@gdf>
2023-02-27 15:01:43:579095 [INFO] Recognition Result: transcript [imię] confidence [0.00] final [0] end-offset [1:800] <ef0dedaf887245f8@gdf>
As you may see, none of the interim results above is final, and all of them have a confidence score of 0. You may need to reach out to Google to understand the cause.
To get better accuracy, Google recommends performing speech transcription using their enhanced model for telephony, whereas you used the default model
2023-02-27 15:01:41:952661 [INFO] Set Model [default] <ef0dedaf887245f8@gdf>
--
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/e8cea192-52c6-41a8-a177-3d70d45e50f1n%40googlegroups.com.
--
Alexander Demidov
Mar 16, 2023, 10:10:51 AM3/16/23
to UniMRCP
Thank you for your reply!
Regarding default model, unfortunately Google doesn't support enhanced mode for pl-PL (based on their docs). Anyway, thank you for highlighting that
Regarding default model, unfortunately Google doesn't support enhanced mode for pl-PL (based on their docs). Anyway, thank you for highlighting that
Vipul Patel
Mar 16, 2023, 1:25:31 PM3/16/23
to uni...@googlegroups.com, Vimal Gupta
@Vimal Gupta here is the reply
To view this discussion on the web visit https://groups.google.com/d/msgid/unimrcp/581c53b8-b45e-4281-8686-12993d76cbf2n%40googlegroups.com.
Reply all
Reply to author
Forward
0 new messages