Skip to content
  • There are no suggestions because the search field is empty.

EMMA Is Not Recognising What a Patient Is Saying

If a patient reports that EMMA did not understand them, or if you notice a submission contains incorrect information that suggests EMMA misheard something, this article helps you understand why it may have happened and what to do about it.

Purpose of This Article

EMMA's speech recognition is built specifically for NHS primary care and improves continuously. However, recognition issues can still occur. This article helps your team understand what is normal, what is not, and when to escalate.

Main Content

How EMMA Listens

EMMA's speech recognition is powered by Clear Voice — a system trained specifically on NHS primary care audio. This means it has been trained on regional accents from across the UK, clinical vocabulary, medication names, and the acoustic challenges of phone calls from various environments including care homes, busy households, and outdoors.

Several features work alongside Clear Voice to improve the patient experience:

  • Pulse — reduces the pause between the patient speaking and EMMA responding, making conversations feel more natural
  • Flow — adapts EMMA's conversational timing to the patient's speaking pace, handles interruptions, and detects when the patient has finished speaking
  • Sentinel — monitors every call and ensures that if EMMA encounters something she cannot reliably interpret, she falls back to a clinically safe pathway rather than guessing

EMMA has processed millions of calls across NHS surgeries and continues to improve with every interaction.

What Can Cause Recognition Issues

Despite continuous improvement, the following factors can contribute to recognition difficulties:

Audio quality

  • Poor mobile phone signal
  • Background noise — a busy kitchen, traffic, a care home environment
  • Speakerphone use, which distorts the caller's voice
  • A low-quality handset

Speech factors

  • Very strong regional accent combined with poor line quality
  • Unusual phrasing for a request type EMMA has less training data for
  • Speaking very quickly or very quietly
  • A patient who pauses frequently mid-sentence

Specific vocabulary

  • Less common medication names or clinical terms
  • Unusual proper nouns such as unfamiliar place names or specialist clinic names

Call conditions

  • Patient calling from an area with weak signal
  • Patient using a VoIP line with compression artefacts

What EMMA Does When She Cannot Understand

EMMA is designed to handle recognition uncertainty gracefully. She does not guess. Instead she will:

  • Ask the patient to repeat themselves or rephrase
  • Offer a confirmation step for ambiguous responses — for example, if a word could be either "prescription" or "subscription", she will ask the patient to confirm
  • Fall back to a clinically safe transfer pathway if she cannot reliably continue the call

Patients are never left stranded. If EMMA genuinely cannot proceed, the call transfers to reception or SOFIA sends an SMS so the request can continue in writing.

What a Recognition Issue Looks Like in Your Inbox

If EMMA misheard something during a call, your team may notice:

  • Incorrect information in a submission — for example, a medication name spelled incorrectly or a condition name that does not match what the patient said
  • A request routed to the wrong category — for example, a clinical enquiry appearing as an administrative submission
  • A submission with unusually sparse or incomplete information

If you notice any of these, review the call recording and transcript in the dashboard to confirm what was actually said.

Step-by-Step Checks

Check 1: Review the call recording and transcript

Log in to the dashboard and locate the relevant call. Listen to the recording and review the transcript to understand what happened during the conversation and whether a recognition issue occurred.

  1. Log in at https://quantumloopai.com/ql-dashboard
  2. Locate the call by date, time, or patient name.
  3. Listen to the recording, paying attention to any points where the patient needed to repeat themselves.
  4. Check the transcript for any words or phrases that appear incorrect.

Check 2: Was the call quality poor?

Listen to the recording and assess the audio quality. If the patient's voice is muffled, broken, or very quiet, poor call quality is the most likely cause of any recognition issue rather than a problem with EMMA.

Poor call quality is usually on the patient's side — their signal, their handset, or their environment. EMMA cannot improve audio that arrives at poor quality. However, she is designed to ask patients to repeat themselves clearly when she cannot hear them properly.

Check 3: Is this a one-off or a pattern?

An isolated recognition issue on a call with poor audio quality is expected occasionally. If you are noticing the same type of recognition error repeatedly — for example, a particular medication name consistently being misheard — this is worth reporting so QuantumLoopAI can review and improve EMMA's handling of that specific term.

Reporting a Recognition Issue

If you believe a recognition issue is causing a genuine problem — incorrect submissions, incorrect routing, or a pattern of errors — report it to the support team.

What to include:

  • Your surgery name and ODS code
  • The date and approximate time of the call
  • A description of what EMMA misheard and what the patient actually said
  • Whether this is an isolated incident or a recurring pattern
  • The call recording reference if available

Contact: support@quantumloopai.com or your account manager.

QuantumLoopAI uses reported recognition issues to improve EMMA's vocabulary and training data. Your feedback directly contributes to ongoing improvement.

Troubleshooting

A patient says EMMA kept asking them to repeat themselves. This is most likely caused by poor audio quality on the patient's line. EMMA asks patients to repeat themselves when she cannot reliably interpret what was said. Advise the patient to call from a location with better signal, or from a landline if possible. If they continue to have difficulty, SOFIA can capture their request in writing after the call drops.

A submission contains incorrect information — EMMA misheard a medication name. Review the call recording to confirm what was said. If EMMA misheard a medication name and submitted an incorrect entry, your team should correct the submission manually before actioning it. Report the specific medication name to support@quantumloopai.com so EMMA's vocabulary can be updated.

A clinical request was routed as an administrative submission. This may indicate a recognition issue or an ambiguous phrase that EMMA interpreted differently than intended. Review the call recording and report the specific phrasing to support@quantumloopai.com. Routing improvements are made in response to reported patterns.

A patient with a strong accent says EMMA could not understand them at all. EMMA's Clear Voice system is trained on regional UK accents and continues to improve. In isolated cases, a combination of strong accent and poor line quality can make recognition difficult. If the patient was not transferred and the call ended without a resolution, this should be reported. Contact support@quantumloopai.com with the call details. SOFIA should have activated if the call dropped.

EMMA seemed to get stuck in a loop asking the same question repeatedly. EMMA has automatic loop detection. If she detects a repetitive pattern, she transfers the call to reception as a safety measure. If a call appeared to loop before transferring, review the recording and report it to support@quantumloopai.com.

FAQs

Does EMMA improve over time for our specific patient population? Yes. EMMA's Learn feature means every call contributes to her ongoing improvement. Additionally, QuantumLoopAI uses reported recognition issues to make targeted improvements to vocabulary and call flow handling.

Can we ask EMMA to slow down for elderly patients? EMMA's Flow feature automatically adapts her conversational timing to match the patient's speaking pace. If a patient speaks slowly, EMMA adjusts accordingly. No configuration is needed.

What should a patient do if they cannot get EMMA to understand them? If a patient is struggling, they can ask to speak to reception. EMMA will attempt to transfer them. Alternatively, if their call drops, SOFIA will send an SMS link so they can complete their request in writing. Patients can also be directed to your online consultation form as an alternative.

Is there anything our practice can do to reduce recognition issues? The main factors are on the patient side — call quality and speaking environment. There is no configuration change that directly improves EMMA's recognition for a specific patient. However, reporting recurring issues helps QuantumLoopAI improve EMMA's handling of specific vocabulary relevant to your practice.