Understanding Fluency

Fluency is a service designed specifically to assess children’s reading fluency. Given an audio file containing children’s speech, and the text of a passage that a child has been asked to read, the service will return an analysis of the audio file in reference to the text passage.

When to Use Fluency

Fluency can be used in formal and informal settings to get an analysis of children’s oral reading fluency.

SoapBox Fluency under the hood

SoapBox Fluency uses a speech-to-text system to transcribe the speech of the child while they read, and a fluency algorithm to perform an analysis on the text output and compare it to the passage that the child was asked to read. These components are described in a bit more detail below.

Speech-to-text system

SoapBox uses its proprietary acoustic models (AM), pronunciation dictionary and custom language model (CLM) to create a speech-to-text system that underpins the Fluency service. Our acoustic models are trained on thousands of hours of child speech data in real world conditions and cover a broad range of accents, dialects and settings, to ensure the most equitable and accurate speech recognition performance on children’s voices.

Custom Language Model (CLM)

A language model is a core part of a speech recognition engine. It is used to transcribe or recognise speech present in an audio file. It contains a large list of words and their probability of occurrence. You can find more information on language modelling here.

For fluency solutions accuracy is important. Most off-the-shelf speech-to-text providers use a generic language model that covers most use cases . This works fine in some scenarios but such off-the-shelf solutions tend not to perform well for children's speech. For this reason we provide the ability for every customer to customise their own solution, effectively creating a bespoke solution with high accuracy for children's speech and your content. 

SoapBox Speech-To-Text and Fluency require a CLM for every client to ensure the optimum results for their use-case. We will work with you to deliver your customised solution in a quick and efficient manner.

Further details around CLMs can be found in the Understanding Custom Language Models section.

Fluency algorithm

The fluency algorithm compares the text read by the child (as transcribed by the speech-to-text component) and the passage that the child was asked to read. The results of the analysis are then returned to the user, including information about when the child omitted or substituted words in the reading passage, when they inserted extra words, or repeated themselves. This information, along with file duratin, start and end time or words and confidence scores can be used to create a words correct per minute (WCPM) or similar metric by which the child’s reading fluency can be expressed.

Technical Docs

Further documentation regarding our Fluency service is available here.