CFD w/Google Speech to Text not working.

[IslandTransit]

Hello,

I have a new installation and working on the main auto attendant via CFD. I have created the google account and related speech and text configuration as documented. In my CFD app, I have configured the json authentication file for google services.

The IVR app runs properly if you use DTMF tones and can traverse through the menus properly, however the Speech to Text is either not storing or processing any returns from Google.

I can see in the log file that the system begins listening to the caller where it needs to be listening and stops when a DTMF tone is received, but no log entry occurs when speaking prior to the timeout and not pressing a key,

I have the Dictionary and Hints set with each of the options.

As a note, I know I'm properly connecting to Google TTS, as the menu is comprised of TTS, however STT doesn't seem to be responding.

I am not seeing any errors listed in the log file to give me a starting point to resolve this either.

 

[edossantos]

Hello,

In the log you sent, every Voice Input component ends the speech recognition because a DTMF input is detected. When you configure your Voice Input component to support both speech recognition and DTMF, then any DTMF will cancel the speech recognition and move on.

Usually there is a short delay since 3CX starts streaming the audio, and the time when we receive the first recognized text from Google. This might be around 2 seconds, depending on what you speak. Please give it some time for this to complete, do not press any DTMF tones during your testing, and check the updated logs.

 

[IslandTransit]

Ernesto,

Yes, I am aware in the log reflects the fact a DTMF tone was generated. This was done after the initial timeout and the let's try again while I was testing movement through the menus. I perhaps didn't clarify this well in my message.

I have tested multiple times repeating my request until the timeout, and then repeating the same or other option request, such as Home office, Remote office, Repeat, etc. with no result until it finally times out. I have also waited for it to time out and then try on the second time through with my verbal requests with no change in behavior.

I have attached a fresh log snip with the call where no buttons are pressed and I repeated my request multiple times throughout the menus and allowed it to retry until it finally hung up (intended).

 

[edossantos]

Just for testing, try configuring a larger Timeout for the Voice Input component, for example 10 seconds. This way you give it more time to Google to return the speech recognition. This is just to understand if the issue is related to the time it takes to return the recognition, or there is something else.

 

[IslandTransit]

Adding extra time to the prompts allowed it to work properly.... sort of.
Apparently the system doesn't begin to listen until all of the announcement is played unlike other systems. Is there a way to get the system to start listening as soon as the announcement begins rather than when complete? Most people are repeat callers and are going to quickly learn what to say for what they want... and will likely get frustrated having to repeat themselves...DTMF appears to work at any point. I'd prefer not to add the "please wait until the announcement is complete before speaking your request" message.

Putting a prompt playback following the CheckRecognition segment and leave the Initial Prompts short or empty didn't seem to play until after the timeout?

Thanks.

 

[edossantos]

The recognition starts when the prompt playback ends, this can't be changed. On the other hand, DTMF digits are processed immediately and they can interrupt the prompt playback. Recurring callers should probably use DTMF digits, actually if you know the option that's always faster and more accurate than speaking.

 

[IslandTransit]

Ernesto,

Thak you so much for getting back. That said, we are using the voice recognition for ADA requirements to support the blind and physically impaired. These are nearly all repeat callers and will quickly learn the menu.

As for a feature request, please add this change in functionality to the list please, as this will be appreciated by the clients we serve.

Cheers,

David