Spelling Post-Processing for Phone

Supported language locales are de (Germany) and en (North America).

This service enhances the accuracy of transcriptions by correcting common misinterpretations of alphanumeric characters spelled out during phone calls.

For an illustrative guide on the Spelling Post-Processing service, watch the provided short video:

https://drive.google.com/file/d/12wRgZy_uCjO__hBGJc-h3A5L6tngFBbZ/view?t=1drive.google.com

Service Configuration

Branches

• success: Indicates successful processing of the input.

Input

• utterance: Retrieve your raw utterance from a storage block using intent.rawPlatformRequestBody.text (phoneV2 only). You can then use this storage variable in the service call.

• language: The language of your input. Currently, German (de) and English (en) are supported.

• numbers: (Optional) Specifies whether mis-transcribed numbers should be corrected. Pass any non-empty value to activate. Note: This parameter is enabled by default for backward compatibility.

• letters: (Optional) Specifies whether mis-transcribed letters should be corrected. Pass any non-empty value to activate. Note: This parameter is enabled by default for backward compatibility.

Output

• result: Your processed input.

• ssml: Your result formatted in SSML, wrapped in <say-as interpret-as="characters">...</say-as>. Spaces are used to indicate groups separated by a short pause.

Technical Configuration

• URL: https://parloaservices.azurewebsites.net/api/SpellingProcessing?clientId=<CLIENT_ID>

• Header: x-functions-key: <AUTHCODE>

For authentication, replace <AUTHCODE> and <CLIENTID> with the values provided by your Parloa representative.

Usage in Parloa

Sample Configurations with Output

Input Configuration in the Frontend

Variables required for service inputs may come from storage variables (for example, lastUtterance) or be directly input as plain text (for example, the language code en).

Block Configuration in the Parloa Frontend

Output

You can use the SSML output to instruct the chatbot on pronouncing each individual letter or number clearly in the dialogue. Alternatively, you may employ the "result" output for further processing and integration of the corrected utterance into the conversation flow.

For de-DE Locale: Several features of the Spelling Processing Service are incorporated into Parloa's specialized STT models. If you're using these STT models in German, the Spelling Processing service may not provide additional enhancements.