Today we are pleased to announce the launch of a new major version of the Speechace API. v9.0 culminates broad investments we have made over the past 18 months in the assessment of spontaneous speech. Everything we learned from creating and launching the Speechace Speaking Test is now available to use as an API with Speechace API v9.0.
What’s new in Speechace API v9.0
1. Overall score and comprehensive subscores
The Speechace Premium API now provides an Overall speaking score and subscores based on evaluating: Relevance, Fluency, Coherence, Pronunciation, Grammar, and Vocabulary.
The scores are returned in 3 formats:
A speechace_score on a scale of 0 to 100
An ielts_score on a standard IELTS scale of 0 to 9.0
A pte_score on a standard PTE scale of 10 to 90
Below is a snippet from the API response. You can refer to the Premium API Docs for REST API signature and example calls.
v9.0 comes with upgraded Scripted Speech and Spontaneous Speech scoring models for evaluating Pronunciation, Relevance, Fluency, Grammar, and Vocabulary. The latest models all incorporate our learnings from the usage of the Speechace Speaking Test as well as core improvements in our acoustic model, alignment and performance.
3. A new API versioning scheme
Up to v9.0 Speechace API versioning was explicit with the caller specifying the exact version they use in the request url.
From v9.0 onwards the API caller can choose to take silent minor version updates or not. For example:
This request will explicitly use the minor version v9.0
4. Added score explainability
v9.0 introduces the score_issue_list, a list that contains any reasons that a particular score was reduced or omitted. For example, attempts which receive a zero overall score due to being irrelevant, too short, or too similar to the question prompt will include a warning and a detailed message explaining the reason.
Below is an example snippet from the API result. You can refer to the API Docs for specific examples:
"detail_message": "The response is not relevant to the context."
v9.0 introduces a number of breaking changes for those upgrading from prior versions. In all cases the deprecated fields were replaced by new fields better organized and integrated with newly added scores.
With this release we round out our scoring models for Spontaneous Speech. Moving forward expect more from us on score feedback and explainability to help learners know where to improve and as always more investments to in scoring and the API platform that supports all of our customers.
You can get started with Speechace API v9.0 today at: