Product updates

In this page, we’ll publish improvements and fixes to Speechly products.

5th of March

Training time estimates

Changed the way how we estimate deployment times when training the models.

Playground

Pressing ‘alt’ temporarily enables the Try-button so that one can use the Playground with an old model while a new one is being trained.

Model deployment

More robust tranining infrastructure that reduces model training times in certain situations.

22th of February 2021

iOS UI Components

Ready-made UI components for microphone and transcript for hastened development on iOS.

Android Client

An Android client for easy integration with Speechly is now available.

The Android client enables quick and efficient development of Speechly applications for native Android applications.

Support for new entities

Support for natural time expressions such as “fifteen thirty”, “20 past nine” or “5 minutes after midnight” with $SPEECHLY.TIME standard variable and entity type Time.

Minor bugfixes

• Better support for long utterances
• Other fixes

8th of February 2021

Debugging models in Speechly CLI tool

New debugging feature in Speechly CLI tool displays example utterances for a given configuration and calculates statistics about occurences of intents and entities.

Support for unadapted ASR

Typically Speechly SLU models are adapted for a specific use case, which helps improve speech recognition accuracy. Now you can also use unadapted ASR for pure transcription use cases. You can test the speech recognition performance here

Support for new entities

Speechly Annotation Language supports natively phone numbers, emails, person names and website addresses. This enables developers to easily build voice experiences that contain these data types, for example something like “Add contact with name Jack Johnston and email address jack dot johnston at gmail dot com”

Minor bugfixes

25th of January 2021

Improved audio handling in browser clients

Streaming audio and handling API results is done by using multiple threads for improved performance. Main UI thread is never blocked by Speechly, resulting in a faster UI.

Minor bug fixes

11th of January 2021

ASR Improvements

New baseline model for speech recognition improves ASR accuracy in all use cases.

Improvements to ASR adaptation

Utterances that are out of domain (ie. no example utterances provided in the model configuration) now have better speech recognition results.