Skip to content
English - United Kingdom
Contact us
  • There are no suggestions because the search field is empty.

EMMA is Mishearing Patient Names Due to Accents or Unusual Spellings

Purpose of Document

Name mishearing complaints are among the most emotive issues raised by practices, particularly in areas with large South Asian, Eastern European, or elderly patient populations. Practices may use this to question whether EMMA is appropriate for their patient demographic. Support staff must respond with accuracy, honesty, and confidence, grounded in the real capabilities of the platform as of V12.

Background: How EMMA Listens and Recognises Names

EMMA uses several interconnected systems to hear, interpret, and verify patient names.

Clear Voice (V12, May 2026)

Clear Voice is EMMA's rebuilt speech recognition layer. It was trained and tuned specifically for NHS primary care audio. This means it has been trained on:

  • Regional UK accents including Yorkshire, Welsh, Scottish, Geordie, West Midlands, and many more
  • South Asian names and accents common across NHS patient populations
  • Eastern European names and phonetic patterns
  • Elderly speech patterns including slower pace, softer volume, and less precise articulation
  • Real-world acoustic conditions: background noise, phone signal variation, calls from busy environments, care homes, and outdoor locations

No other AI reception platform has trained at this scale on NHS-specific audio. Clear Voice represents a significant improvement over the speech recognition used in previous EMMA versions.

Pulse (V12, May 2026)

Pulse is EMMA's response speed layer. It reduces the pause between a patient speaking and EMMA responding. Crucially for name recognition, Pulse allows EMMA to wait a natural amount of time before processing, reducing false starts and cut-offs when patients say their name slowly or with pauses.

Flow (V12, May 2026)

Flow is EMMA's conversational timing layer. It reads each caller's speaking rhythm and adapts dynamically. For elderly patients who speak slowly and carefully, Flow adjusts EMMA's pace to match. Per-practice demographic tuning means Flow adapts not just to each individual caller but to the communities a specific surgery serves. A surgery in a predominantly elderly area can have Flow configured to lean towards a slower, more patient conversational style.

Sentinel (V12, May 2026)

When EMMA encounters something genuinely unfamiliar, including a name she cannot confidently interpret, Sentinel ensures she falls back to a clinically safe pathway rather than guessing. This means EMMA will ask the patient to clarify or spell their name rather than submit an incorrect name on a form.

Easy Confirm (V12, May 2026)

Easy Confirm progressively reduces repetition in how EMMA confirms patient inputs. Over a four-week rollout, EMMA is reducing unnecessary repeated confirmations while maintaining clinical safety thresholds. This makes the name collection step faster and less frustrating for patients who have unusual names and have had to repeat or spell them multiple times.

What EMMA Does When She Cannot Hear a Name Clearly

When EMMA cannot confidently interpret a patient's name, she does not guess or submit an incorrect entry. Instead she follows a structured fallback sequence:

  1. EMMA asks the patient to repeat their name
  2. If she still cannot interpret it confidently, EMMA asks the patient to spell their name letter by letter
  3. EMMA reads back what she has recorded and asks the patient to confirm or correct it
  4. As of 1 May 2026, the read-back is delivered clearly and at a measured pace, making it easier for patients to catch errors

This is the same process a human receptionist would follow. It is not a failure. It is a safe and appropriate fallback.

Regarding the phonetic alphabet: EMMA does not ask patients to use the phonetic alphabet. If spelling is needed, she asks patients to go letter by letter in plain language. Known edge cases exist where letters that sound similar such as S and F can still occasionally be confused. These are actively being trained on and improve with every call EMMA processes.

Common Causes and What to Check

1. A patient has a South Asian name that EMMA struggled to recognise

This is the most commonly reported accent-related complaint. South Asian names span a wide range of phonetic patterns and regional variations that historically presented challenges for generic speech recognition systems.

What to do:

  • Review the call recording to assess how EMMA handled the name
  • Confirm whether EMMA asked for the name to be spelled and whether that process worked
  • Check whether the surgery is on V12 with Clear Voice active. Clear Voice has specifically trained on South Asian names and NHS patient demographics.
  • If the surgery is on V12 and persistent issues are occurring with a specific patient demographic, raise a demographic tuning request with engineering

2. An elderly patient speaks softly or slowly and EMMA struggled to keep up

Elderly patients frequently speak more quietly, more slowly, and with less precise articulation than younger callers. Before V12, this could cause EMMA to mishear or cut off before the patient had finished speaking.

What to do:

  • Confirm whether the surgery is on V12 with Flow and Pulse active
  • Flow adapts to the patient's speaking pace dynamically, and Pulse reduces premature cut-offs
  • If the surgery is on V12 and elderly callers are still experiencing recognition issues consistently, raise a per-practice demographic tuning request to adjust Flow for that surgery's patient population
  • Consider recommending SOPHIA as a preferred channel for patients who find speaking to an AI challenging. SOPHIA allows patients to interact by typing, which removes the speech recognition barrier entirely. Patient feedback on SOPHIA has been overwhelmingly positive.

3. A patient with a strong regional accent was consistently mishearing

What to do:

  • Review the call recording to assess whether the accent appears to be the primary cause
  • Confirm whether the surgery is on V12 with Clear Voice active
  • If persistent issues occur with a specific regional accent at a specific surgery, raise a demographic tuning request
  • Note the specific accent pattern and any specific names or words that are being consistently mishearing, as this information will help the engineering team prioritise the tuning

4. Background noise on the call made name recognition unreliable

Calls made from a busy kitchen, a bus, a care home corridor, or outdoors can introduce significant background noise that affects any speech recognition system regardless of how well trained it is.

What to do:

  • Review the call recording and confirm whether background noise is clearly audible
  • If background noise is the primary cause, this is not a platform fault
  • If the patient was on a mobile and the call dropped or became very unclear, confirm whether SOPHIA sent an SMS continuation link so the patient could complete their request in writing
  • SOPHIA is a meaningful accessibility win for patients who struggle with voice-based interaction in any environment

5. EMMA asked the patient to spell their name and the process did not go smoothly

When EMMA asks a patient to spell their name, occasional letters can be confused, particularly S and F which sound similar. This is a known edge case being actively trained on.

What to do:

  • Review the call recording to assess where the spelling process broke down
  • Confirm whether EMMA ultimately succeeded in capturing the correct name through the spelling and read-back process
  • If EMMA captured the wrong name despite the spelling process, note the specific letters that were confused and raise a flag to engineering
  • If this is happening repeatedly for patients with names containing specific letter combinations at a particular surgery, raise a tuning request

SOPHIA as an Accessibility Alternative

For patients who consistently find speaking to EMMA challenging due to accent, hearing impairment, speech impediment, or simply a preference for typing, SOPHIA is a meaningful and positive alternative.

SOPHIA allows patients to interact via text on their mobile device or computer. There is no speech recognition involved. The patient types their name, date of birth, and request at their own pace. The practice receives the same structured outcome.

As of V12, SOPHIA is described as an accessibility win. Patient feedback on SOPHIA has been overwhelmingly positive, and for many callers it is now the preferred channel.

Support staff should be comfortable recommending SOPHIA to practices whose patient populations include groups who may find voice interaction more difficult.

How to Raise a Demographic Tuning Request

If a surgery is experiencing persistent name mishearing for a specific patient demographic or accent pattern, a demographic tuning request can be raised with the engineering team.

To raise a tuning request:

  1. Confirm the surgery name and ODS code
  2. Identify the specific patient demographic or accent pattern affected (for example: Punjabi names, Welsh names, elderly patients with slower speech, strong Geordie accent)
  3. Provide specific examples from call recordings where the mishearing occurred, including call IDs and the names that were misheard versus what the patient actually said
  4. Note any specific letters or phonetic patterns that appear to be causing the confusion
  5. Raise the request with engineering, flagging it as a demographic tuning request for Clear Voice or Flow as appropriate
  6. Set the expectation with the practice that tuning improvements are applied iteratively and may take a short period to fully reflect in call quality

Step-by-Step Triage Process

  1. Ask the practice to describe the specific complaint. Which patients are affected? What names are being mishearing? What is appearing on the form versus what the patient said?
  2. Review the call recording. Confirm whether EMMA asked for spelling and how that process went.
  3. Confirm whether the surgery is on V12 with Clear Voice, Flow, and Pulse active.
  4. Assess whether the issue is accent-related, background noise related, or an edge case with specific letter sounds.
  5. If the surgery is on V12 and persistent issues are occurring with a specific demographic or accent pattern, raise a demographic tuning request.
  6. If the surgery is not yet on V12, flag that V12 includes significant speech recognition improvements and the issues may improve post-migration.
  7. If EMMA is capturing incorrect names despite the read-back step being completed correctly, escalate to engineering.

What to Tell the Practice

Keep communication honest, factual, and reassuring. Do not overpromise perfection.

Example wording for a practice with a large South Asian patient population:

"EMMA's speech recognition has been trained specifically on NHS primary care audio, including a wide range of South Asian names and accents that are common across NHS patient populations. With our V12 release, this training has been significantly expanded. For patients whose names are particularly unusual or complex, EMMA will ask them to spell their name letter by letter and then reads it back for confirmation. If you are seeing recurring issues with specific names, we can look at demographic tuning for your surgery to improve accuracy further."

Example wording for a practice with a large elderly patient population:

"EMMA's V12 release includes a feature called Flow which adapts to each patient's speaking pace. For elderly patients who speak more slowly or quietly, EMMA adjusts her rhythm to match. We can also configure this specifically for your surgery's patient population. For any patients who find speaking to EMMA challenging, our web assistant SOPHIA allows patients to type their request instead, which removes the speech recognition step entirely and has received very positive feedback from patients across all age groups."

Common Mistakes

  • Telling a practice that EMMA uses the phonetic alphabet. She does not. She asks patients to spell letter by letter in plain language.
  • Dismissing accent-related complaints without reviewing the call recording.
  • Not mentioning SOPHIA as an alternative for patients who struggle with voice interaction.
  • Promising that name recognition will be perfect after a tuning request. Tuning improves accuracy incrementally. Set realistic expectations.
  • Not checking whether the surgery is on V12 before advising on speech recognition capabilities. Pre-V12 and post-V12 behaviour differ significantly.
  • Treating every mishearing as a platform fault without considering background noise or patient-side acoustic conditions.

Escalation Guidance

Escalate to engineering if:

  • EMMA is consistently mishearing names after V12 Clear Voice is confirmed active, and a demographic tuning request has already been raised without improvement
  • EMMA is capturing incorrect names despite the spelling and read-back process completing correctly
  • A clinical safety concern has been raised because an incorrect name appeared on a submitted form and was actioned without being caught

When escalating, always include:

  • Surgery name and ODS code
  • Specific patient demographic or accent pattern affected
  • Call IDs of affected calls
  • Names that were mishearing versus what the patient actually said
  • Specific letters or phonetic patterns causing confusion if identified
  • Whether a demographic tuning request has already been raised and when

Last Reviewed: May 2026 Owner: Support and Customer Success