Improve your transcripts

By selecting the Help improve your transcriptions checkbox in your calling settings, you are giving HubSpot the permission to store, manually transcribe, and label audio from calls. Providing access to your call audio means HubSpot will be able to improve speech-to-text capabilities based on how you speak. As part of this improvement process, HubSpot may manually transcribe and take notes on the shared audio.

Please note: it will take time for this data to be used in HubSpot's processes. Expect transcripts to improve over time.

Security & privacy

Recordings received by HubSpot may include personal data, and may be reviewed by human reviewers and transcribed into text for HubSpot’s product, services, and improvement efforts. HubSpot works with trusted third party vendors to help improve these services. The vendor will only have access to your recording during the transcription process.

All data is shared, processed, and stored in accordance with HubSpot’s Privacy Policy.

Why does HubSpot need this data?

Improve performance

HubSpot uses the account specific data to measure and improve the machine learning models that power Conversation Intelligence (CI).

HubSpot measures transcription quality across participating users. Any number of factors can contribute to transcript quality. It is important to have a representative sample of industries, languages, and accents to make sure HubSpot can provide a quality service across these different variables.

Before audio can be used in training or for benchmarking HubSpot must get a manual transcript. These manual transcriptions can then be labeled to provide further understanding to the models. For example, HubSpot can detect words that are important to your company, questions asked, or conversation breakdowns like small talk, next steps, and pricing. This understanding provides the labels that power HubSpot's CI service.

Provide a better customer experience

Higher quality speech-to-text services will ensure that all CI features like keyword search, tracked terms, and other models like call topic tagging work well for you.

Having a representative sample is important so the model can learn from edge cases that are specific to your company. Anything from how noisy your office is, to industry specific terminology, to accents can affect the model. By allowing HubSpot to sample your audio, HubSpot can take these edge cases into account. 

Automatic transcription is still new. Check out this article from Descript for more information.