Can AVT understand different accents and non-native English speakers?

“Will it understand my accent?” was one of the questions we heard more than once at the Voice Technologies User Conference in September 2026, along with a closely related concern: what if English isn't my first language?

These are important questions. Ambient Voice Technology (AVT) is designed to work with natural spoken language, but speech recognition accuracy can vary when clinicians and patients have different regional accents or speak English as an additional language.

Why accents can be challenging for speech recognition

People speak English in an enormous variety of ways. Regional accents affect pronunciation, while non-native English speakers may bring different sounds, rhythms and speech patterns to a conversation.

Independent research confirms that these variations can affect speech recognition performance. A 2026 study published in npj Digital Medicine tested general-purpose Whisper and WhisperX models using clinical speech from native and non-native English speakers. Both produced more errors for non-native speakers, although adding AI language-model processing substantially reduced the disparity. Read the study in npj Digital Medicine.

The important question isn't simply whether AI can recognise speech, but whether the technology has been designed and tested for the variety and complexity of real clinical conversations.

Why healthcare-specific speech recognition matters

AIVA (AI Voice Assistant), embedded within Voice Technologies' applications, uses Corti's clinical-grade AI, developed specifically for healthcare rather than general-purpose transcription.

Corti explains that training and fine-tuning speech recognition models should account for real-world speaker characteristics, including accent, dialect, background noise and recording equipment. Its models can also be tuned for particular dialects where required.

This approach to accent recognition is particularly important in healthcare, where accurate clinical documentation depends on understanding diverse voices and speaking styles. Read more about Corti's approach to speech recognition training.

Clinical speech recognition also needs to understand specialist medical terminology, including drug names, conditions, measurements and abbreviations. Corti's Symphony speech-to-text technology is designed for clinical language and reports 98.6% word accuracy in its English real-time benchmark, alongside specialist medical term recognition. Explore Corti's clinical speech-to-text technology.

Designed for real clinical conversations

AVT doesn't require clinicians to dictate carefully constructed sentences - it captures natural conversations between clinicians and patients, making speaker variation, regional accents and different speech patterns important considerations.

However, clinical speech-to-text accuracy is not identical across every speaker or environment. NHS organisations should evaluate how AVT performs with real-world clinical speech, including different accents, non-native English speakers, background noise and specialist clinical terminology.

As with any AI-generated clinical notes, the clinician remains in control. Notes should always be reviewed and approved before becoming part of the clinical record.

Voice Technologies' healthcare voice technology solutions combine specialist speech recognition with established clinical workflows.

For a broader explanation of the technology and its role within NHS clinical workflows, read AVT in healthcare: a complete guide to AI clinical documentation for NHS and GP teams.