This is article overviews a two-part workflow setup: one to automate transcription after a recording has been made, and the second to automate exporting transcriptions via an API request.

User Level:

Admin

Speech-to-Text Two Part Setup

Before Starting

In order to access the features needed for this setup, make sure:

You've successfully been added to an Expert or Custom plan with CloudTalk
CloudTalk Support has enabled Speech-to-Text for your account

Billing for speech-to-text depends on whether you choose to use an enhanced model for transcription, which is an option for some languages. For an enhanced transcription, the rate is 0.072 per minute, while for the default model, the rate is 0.048 per minute.

Transcribe Call Recordings

This section of the article focuses on creating the first workflow automation to handle Speech-to-Text transcription creation.

Steps

From your dashboard, navigate to Account -> Workflow Automations.
Select +Add Workflow and name your workflow.
As a trigger, select recording uploaded.
1. uploaded is the only available Action type for a recording Object.
2. Note: If no conditions are set, this automation will run for every recording that gets uploaded.
Optional: Add conditions if you want transcribe recordings only in cases where these conditions are met.
Select the Action Speech to Text.
1. For Key: language, enter the appropriate language reference from the BCP-47 column in Google Cloud's language list.
  1. Note: While we support all languages listed, not all have an Enhanced Phone call model available.
2. For Key: model, you can choose a specific machine learning model. In most cases, phone_call is best. If having issues, try default.
3. Set Key: enhancedModel to true to use the advanced phone call model, which optimizes for accuracy.
  1. Supported languages for this option are: English (AU, GB, US), French (CA, FR), German, Italian, Japanese, Portuguese, Russian, and Spanish (US, ES).
  2. Using this model increases the costs of transcription.
Confirm and continue.
Save your workflow and activate it by toggling the switch at the top of the workflow setup page.

Forward Transcriptions to external API

This section of the article focuses on setting up the second workflow automation to handle the export process.

From your dashboard, navigate to Account -> Workflow Automations.
Select +Add Workflow and name your workflow.
Define a trigger of Object type transcription and Action type created. This trigger will initiate any time a transcription gets created.
We could choose to set up an optional condition. Choose a dynamic Property and reference its example Value by looking at the Useful data for your Workflow section, if Conditions are necessary. For the purposes of our example
Select the Action API request.
1. Enter the Endpoint URL. This can usually be located in the API docs or within developer mode of the respective service (unless using a custom application).
  1. Example: https://api.hubapi.com/engagements/v1/engagements (HubSpot engagements)
2. Method: POST.
In the Values section (the request body), include the following key-value pair:
1. For Key: transcription, paste the Value:
  {{ event.properties.transcription.csv }}
  1. This is a dynamic value which references the transcription to be exported, converted to a CSV format.
Include any required API request components in both the Values and Headers sections of the API Request configuration, following the documentation for the external API you are using to export your Speech-to-Text data.
Confirm and continue. Save your workflow and activate it by toggling the switch at the top of the workflow setup page.

Troubleshooting Tips

For information regarding transcription, refer to Google's list of languages, transcription models, and advanced transcription models.
In most cases, sending information to another API (such as one associated with a CRM or other internet-connected tool) requires one or more keys or authorization tokens. Ask an admin familiar with the receiving API if you have trouble finding this information.

Faulty conditions are often the result of unrecognized inputs in the Property box, or inputs in the Value box which will always result in no matches.
After clicking into the Property box, you must click one of the pre-selected keywords within the Useful data for your Workflow section to the right.
You can also refer to the Useful data for your Workflow section to validate whether a Value you entered seems to be a good match for a property you've chosen.