Tool Tuesday: Transcribe with AI

This past ASHA Convention in Seattle, my friends and colleagues Meghan Graham and Caroline Brinkert from Boston University discussed the importance of language sampling in supporting student growth, specifically in preschool. They described several values within language sampling:

• Spontaneous and naturalistic
• Identifying functional areas of need
• Multi-dimensional for Macro- and Microstructure Assessment, treatment planning AND progress monitoring

Additionally, Meghan and Caroline described barriers to language sample analysis, including time investment and clinicians’ uncertainty and lack of confidence in their skills for this kind of assessment. Truly, the time factor was always a big deal. Recording, playing, and rewinding cassette tape recordings, as we typed out a sample, gave way to doing the same with digital recordings on our phones and iPads, saving little time. Speech-to-text tools such as Google Docs’ Voice Typing helped a bit with the pain of transcription, but it was still tedious.

With the recent arrival of easy-to-use artificial intelligence (AI), obtaining a transcription of a language sample is SO much easier! AI transcription utilizes Automatic Speech Recognition (ASR) technology, which is based on language and learning models that interpret human speech and convert a recording into text. There are two good options available for this:

Rev.ai is a HIPAA-compliant site, should that be important to you, and allows for up to 5 hours of transcription (only two cents a minute after that allotment is used). Just go to the Speech-to-Text tab and submit your digitally recorded file as a “job” (a Voice Memos file works). After it processes you can download a TXT file which can be opened in a word processor.

Otter.ai allows for 3 uploaded recordings for free, but you can record live with the website or app and it will transcribe the “meeting,” also good for collecting language samples. Otter’s interface is a bit more user-friendly than Rev’s. Working in private practice, I like Rev’s HIPAA compliance just to be safe, even though it is unlikely that protected health information (PHI), such as client name or birthday, is protected in any of my recordings.

After obtaining the transcript, you’ll want to compare it with your recording to check for any errors in transcription, as AI may be tripped up by a variety of factors: background noise, overlapping speech, homophones, speech sound errors. And, it may or may not transcribe filler words or mazes (Rev.ai is better with this) that may be important to characterize how your student produces narrative. SALT provides a good guide to segmenting any sample into C units to view sentence complexity, and of course, MindWing has made available a variety of tools for analyzing language samples, including their Data Collection and Progress Monitoring manual.

I hope these tools provide you some ease and comfort for initial or ongoing assessment of your students’ narrative language!