Adjusting parameters on the UMSazureSR plugin

[Fidel Gil]

Hello,

Hello I am using MRCP server 1.7.0 with the Azure SR plugin, I have installed and run the diagnostic test of using the bsr1 scenario from the umc-addons.

I have created a copy of the bsr1 scenario and changed it to use a custom audio file. And here comes my problem.

When I set the audio file to be a small audio file, say a single sentence, I get the transcript of it as expected. When I set the audio File to be something larger in my case a 20 second recording, the server gets as reply the following:

MRCP/2.0 243 RECOGNITION-COMPLETE 1 COMPLETE
Channel-Identifier: f56e798f646c4b7f@speechrecog
Completion-Cause: 001 no-match

While on the umc side I get the following output:

[WARN]   No NLSML data available

I have modified the umsazuresr.xml file to not apply single utterances as can be seen here:

   <ws-streaming-recognition
      language="de-DE"
      max-alternatives="1"
      alternatives-below-threshold="false"
      sort-alternatives="false"
      confidence-format="auto"
      confidence-precision="2"
      results-indent="2"
      results-format="standard"
      tag-format="default"
      input-format="default"
      start-of-input="service-originated"
      skip-unsupported-grammars="true"
      transcription-grammar="transcribe"
      grammar-param-separator=";"
      auth-validation-period="480"
      auth-request-timeout="30"
      inter-result-timeout="0"
      input-token="Lexical"

      instance-token="ITN"
      connect-timeout="0"
      auto-reconnect="false"
      max-audio-data-chunks="0"
      max-connection-duration="0"
      graceful-ws-close="false"
      single-utterance="false"
   />

Asides of that while reading the usage guide:
 https://docs.unispeech.io/en/ums/ms-azure/speech-recog/usage

I read that the endpoint could be part of the issue so I changed it from interactive to dicatation. So I specified the endpoint on the cognitive.subscription.key file 

"service-endpoint": "https://germanywestcentral.stt.speech.microsoft.com/speech/recognition/dictation/cognitiveservices/v1"

If anyone could point me on the right direction on how to do the speech recognition in continious mode, that would be awesome. Also I am not well versed on the UNImrcp Server logic so it could be that there is some default I have missed 

Kind Regards

Fidel Gil

[Arsen Chaloyan]

Hi Gil,

Setting the single-utterance attribute to false in the configuration shall be sufficient, unless the parameter is overridden in the recognition request. The server logs located in the /opt/unimrcp/log directory with the sample utterance that you use would be required to identify the problem.

--
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/5d604d9b-2423-48aa-940d-7f4dc4b2229en%40googlegroups.com.

--

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