Speech to text app

[Chris Walters]

I've looked over this guide here and looked at the product demo included in the installation.

I have a slightly different request from a client, this is what they want:

1. When a call comes in, the caller is prompted to speak a description of their issue.
2. The audio description is processed through Google STT API
3. The returned text is sent via email to a specified address along with CID of caller, timestamp, dialed DID
4. The call is then routed to their on-call ring group.

Has anyone had any success with something similar, or is the CFD only able to use matching statements for return from the STT API?

 

[edossantos]

I've looked over this guide here and looked at the product demo included in the installation.

I have a slightly different request from a client, this is what they want:

1. When a call comes in, the caller is prompted to speak a description of their issue.
2. The audio description is processed through Google STT API
3. The returned text is sent via email to a specified address along with CID of caller, timestamp, dialed DID
4. The call is then routed to their on-call ring group.

Has anyone had any success with something similar, or is the CFD only able to use matching statements for return from the STT API?

This is totally doable with the CFD. The Voice Input component lets you recognize free text, and get it in the RecognizedText property after execution. If you define Dictionary entries, then the component will try to match what the user says to an entry. But if the Dictionary is empty, any text can be spoken, and you will get back what Google recognizes. This can be then sent via email, as you need.

 

[Evolute IT]

I've looked over this guide here and looked at the product demo included in the installation.

I have a slightly different request from a client, this is what they want:

1. When a call comes in, the caller is prompted to speak a description of their issue.
2. The audio description is processed through Google STT API
3. The returned text is sent via email to a specified address along with CID of caller, timestamp, dialed DID
4. The call is then routed to their on-call ring group.

Has anyone had any success with something similar, or is the CFD only able to use matching statements for return from the STT API?

Don't set any hint or dictionary. Then, use the RecognizedText result property to get the full transcoded text.

https://www.3cx.com/docs/manual/cfd-components/#h.lgzfai7rclby

Then, just use the Send Email component and the Transfer to send the call to the ring group.

We can build that for you pretty quickly if you prefer.

 

[fxbastler]

Question: how can I get the original recorded file?

 

[edossantos]

Question: how can I get the original recorded file?

You can't at the moment, but this is something that we're planning to add in a future update of the CFD.

 

[Evolute IT]

You can't at the moment, but this is something that we're planning to add in a future update of the CFD.

Nice!

Question: how can I get the original recorded file?

In the meantime, you can still use the recognized text so that they at least have a text version of it.

Or, you can always use the "Record and Email" feature. https://www.3cx.com/docs/manual/cfd-components/#h.ql40cxor0qax

 

[fxbastler]

thx

Record an Email do not work in this scenario. Record, transcribe, do something with the output and then send it all together is not working with the standard CFD blocks. Voice Input without dictionary is the nearest thing.

Still looking at the recording file ...

 

[edossantos]

Still looking at the recording file ...

The audio stream is currently discarded, so you will need to wait for an update from 3CX. Be sure that this will be available in the near future, as we're already working on this.