Fabric scores spoken English on a 0–10 scale using a live conversation, with five dimensions and timestamped evidence from the candidate’s own words. For BPO hiring teams, the score is a structured way to compare communication skills across a batch. It is not a certificate, an IELTS result, or a substitute for checking whether someone can do the job. The scale is calibrated against CEFR descriptors, but its CEFR mapping is approximate. Accent does not change the score; mother tongue influence is described separately and is not scored.
How one score is built
One live conversation,
five things scored
What a recruiter receives for each candidate. Illustration of the structure only: the bars are not a real candidate's scores.
Why spoken-English scoring matters in BPO hiring
When a contact centre hires dozens or hundreds of voice-process agents, recruiters need a consistent way to decide who should move to the next stage. A short phone screen can help, but different recruiters may ask different questions, tolerate different mistakes, or interpret the same accent differently. A score without evidence creates another problem: a hiring manager may see a number without knowing what the candidate actually said.
A useful assessment should make three things visible: what communication skills were evaluated, which parts of the conversation support the score, and how the result relates to the demands of the role. It should also separate English proficiency from accent and from unrelated factors such as whether the candidate gave the “right” answer to a question.
How Fabric’s 0–10 spoken-English score works
Fabric evaluates five dimensions of spoken communication: intelligibility, fluency, vocabulary, grammar, and coherence. Each dimension receives a score out of 10, and the overall score is the average of those five dimension scores. A score of 6 is intended to represent functional professional communication: not flawless English, and not an automatic pass for every job.
- Intelligibility: how readily a listener can understand the candidate’s speech.
- Fluency: how smoothly the candidate speaks, including pauses that indicate difficulty finding words.
- Vocabulary: whether the candidate can choose words that express the intended meaning.
- Grammar: how accurately the candidate forms sentences and uses grammatical structures.
- Coherence: whether ideas connect clearly and are easy to follow.
What each dimension sounds like
Pick a dimension to hear it the way a trainer would.
Can every word be understood?
What Fabric listens for: Articulation, word stress, and whether the ends of words come through.
Does the speech flow?
What Fabric listens for: Pace, where the pauses land, and how often there are restarts, repetitions and filler words.
Are the words precise?
What Fabric listens for: Whether the candidate has the words for everyday professional conversation, and what happens when a word does not come.
Do sentences come out right?
What Fabric listens for: Agreement, tense, articles and prepositions, and how often longer sentences come out intact.
Is the answer easy to follow?
What Fabric listens for: Connectors and signposting, and whether long answers stay easy to follow.
The overall score should not be read in isolation. Two candidates can have the same average but different strengths: one may be easy to understand but struggle with grammar, while another may use accurate grammar but find it hard to organise an explanation. Looking at the five dimensions and the evidence behind them gives recruiters a more useful picture than a single number.
What do the score bands mean?
The bands below are practical interpretations of Fabric’s scale. They are not separate qualifications, and the boundaries should not be treated as universal hiring rules.
Fabric's 0 to 10 scale, and where it roughly sits on CEFR
Read the band from the score, then use the CEFR and IELTS column for orientation only.
These are approximate comparisons, not certified CEFR levels or IELTS-equivalent scores. Fabric's score is its own assessment scale. Distribution: Fabric production interviews, as published in the communication analysis guide.
The score bands overlap at some boundaries because they are intended to help interpret a continuous scale, not create hard linguistic categories. Use the underlying score, dimension breakdown, and interview evidence together. A hiring team should decide its threshold based on the role and its own training and quality standards. Expect most of a typical pool to land between 5 and 8: in Fabric’s production interviews, about half of candidates score 5 to 6.5 and only about 3% score above 8.
How the scale relates to CEFR, and what it does not mean
Fabric’s scale was calibrated using descriptors from the Council of Europe’s Common European Framework of Reference for Languages (CEFR) and checked against the EF English Proficiency Index 2025. That helps provide an external frame of reference for interpreting results. It does not mean that Fabric issues a CEFR result or that a 0–10 score can be converted precisely into an IELTS band.
In a comparison with the public Cambridge Speak & Improve learner dataset, Fabric’s scores had a rank correlation of 0.77; 86% of speakers were within one CEFR level of an examiner’s assessment. Fabric’s scores were about 0.7 points stricter on average. This is useful evidence, but it has a limitation: the dataset contains public language-learning recordings, not BPO customer-service calls. It should not be presented as a direct validation on contact-centre applicants.
Fabric's scores against Cambridge examiners
Fabric's scorer run over the Speak & Improve Corpus 2025, learner recordings with examiner-assigned CEFR levels.
The practical takeaway is to use the CEFR comparison as orientation, then calibrate the pass mark against the communication demands of your own roles. Do not tell candidates that the score is “CEFR certified,” “IELTS equivalent,” or equivalent to a Versant or SVAR result. The assessments use different methods, and no same-candidate comparison between Fabric and Versant/SVAR is being claimed here.
Fabric scores intelligibility, not accent
A candidate’s accent should not lower their spoken-English score. The assessment focuses on whether their speech can be understood and how effectively they communicate, not whether they sound like a speaker from a particular country or region. Familiarity with one accent should not be mistaken for stronger English.
Fabric reports mother tongue influence (MTI) separately from the score. The MTI read can describe the likely accent family, sound substitutions, rhythm or stress carryover, and the strength of those patterns. It is descriptive information for a recruiter or trainer; it is not a score and does not change the overall result.
Two reports from the same recording
Accent is described in one place and scored in none.
Intelligibility
Part of the 0 to 10 score
- Can the listener catch every word the first time?
- Do word stress and word endings come through?
- Does anything force the listener to ask again?
MTI read
Descriptive, for recruiters and trainers
- Likely accent family
- Typical sound substitutions
- Rhythm or stress carried over from the first language
- How strong those patterns are
For a BPO team, this distinction matters. If a customer can understand the agent clearly, a regional accent by itself should not be a reason to reject them. If a sound pattern regularly makes a key word difficult to understand, the relevant question is intelligibility in the context of the role, not whether the accent is familiar to the evaluator.
What does and does not affect the score?
What moves the English score, and what never does
| Considered in the English score | Not a reason to raise or lower it |
|---|---|
| Intelligibility, fluency, vocabulary, grammar and coherence | Accent or mother tongue influence by itself |
| Pauses that indicate word-searching difficulty | Thinking time by itself |
| How clearly the candidate expresses and connects ideas | Whether the answer is correct or technically deep |
| Language used in the candidate’s spoken response | Answer length, jargon, or enthusiasm |
| Speech that can be assessed in the recording | Background noise, audio quality, or asking the interviewer to repeat a question |
This separation is important because English proficiency is not the same as job knowledge, confidence, personality, or call-centre experience. Evaluate those separately using job-related criteria. If a recording is too noisy to assess reliably, do not treat the noise as evidence of poor English; arrange a better recording or another assessment.
Hear five score bands in real Fabric interviews
A score is easier to interpret when you can hear the speech behind it. Each clip below comes from a real Fabric AI interview and is one of the labelled samples in Fabric’s communication analysis guide. The questions are general opening questions (the AI interviewer asks candidates to share something about themselves, then follows up), not BPO customer calls, so listen for how each person speaks rather than what they talk about. Each clip shows its score band, the approximate CEFR level for that band, Fabric’s description of how the band sounds, and a transcript.
Five score bands, five real candidates
Clips from Fabric AI interviews, labelled with the band Fabric scored. Listen for how the person speaks, not what they talk about.
Precise and easy to listen to across the board, with crisp articulation, complex sentences that stay under control, and clear signposting. What separates this from 9+ is the occasional visible seam, like a restart or a small slip.
Read the transcript (1:45)
Yeah, so I'm actually a petroleum engineering graduate. I studied oil and gas engineering, drilling, reservoir engineering, production engineering, but never really worked in the oil and gas industry. I quickly transitioned into the AI/ML industry and I've been doing, you know, data and AI work pretty much for my entire career, the last nine years. So, I am a kind of person who does not like to think and stick inside a box. I do not, you know, take no for an answer or shy away from taking risks. Hence, that is a pretty good summation of, you know, that my career arc itself is a pretty good summation of me as a personality.
AI interviewer: Petroleum engineering to AI and machine learning is quite a switch. What led you to make that jump away from oil and gas into tech?
Yeah, I mean, honestly, when I was graduating, the industry was going through a downturn and it was kind of hard for me to land a job that paid well and made sense for me. That was one of the reasons, and, you know, considering the amount of effort that it would have taken for me to go into the oil and gas industry, I was not particularly excited about the prospects of working in a remote location, you know, with oil field and, you know, handling heavy...
Transcript of the clip. The candidate's wording is kept as spoken; a place name is replaced with [city]. Interviewer lines in italics.
Comfortably professional. Everything gets across, but one or two habits are noticeable, maybe filler chains, small recurring grammar slips, or uneven pacing. You register them without ever being slowed down.
Read the transcript (2:00)
...about me and make it unique or surprising and share it in an interesting way, because... so I would like to tell about my college days. So it was third year and I and one of my friend who decided that rather than just going and like dancing in the night and like enjoying in the [unclear], we decided that let's try to do some sort of experiment and just try to like get some monetary benefits out of it. So unlike all the students who are excited for like singing, dancing or going meeting celebrities or like enjoying good food, what I decided was that, hey, why not just open a food stall in the college and just try to earn some sort of money from our friends. So what we did was that we opened a food stall in our college fest and we did some, like, it was very interesting, 'cause we tried to get few chefs from nearby [city] and then we tried to price the product, we did some testing also, we sourced the materials and we bought the stall and eventually we did three years of sale and like we didn't made much profit from it but we eventually got to the break-even point. So it was very fun and we learned a lot of things and it was genuinely more exciting than you can hear.
AI interviewer: You mentioned bringing in chefs from nearby. How did you actually convince them to join a college fest food stall?
No, it was not convincing them. Obviously we paid them, so there is no sort of convincing anyone, we just paid them the fair price, but it was since we were college...
Transcript of the clip. The candidate's wording is kept as spoken; a place name is replaced with [city]. Interviewer lines in italics.
Still solidly professional, just with the rough edges spread a bit wider. Both the delivery and the structure show some wear, but you never have to work to follow the meaning.
Read the transcript (2:00)
...fun fact about me is no one knows that, but I really do poetry, but I generally tell them that I do open mics and they seem to guess that I do stand-up comedy. But yeah, I do poetry instead of that.
AI interviewer: So you write poetry, but people think you do stand-up comedy. What kind of poetry do you usually write? Can you share a bit about your style or favorite topics?
I don't have a particular style or a favorite topic. I prefer to write open thoughts. I generally write on other people's thoughts instead of writing on my own. I find it very difficult to write it on my own thoughts. So I prefer writing on other people's thoughts. That's easier for me to process or express.
AI interviewer: Interesting approach. Writing from others' perspectives. How do you usually gather these thoughts from people? Do you ask them directly or pick them up from conversations?
Yeah, generally I'm picking those from the conversations that we had. Like I'm talking to some person, they are sharing something, a certain feeling or a certain movement. And I picked those feelings and moments and I write about how they might be feeling in that particular scenario or, you know, how one should feel in those scenarios.
AI interviewer: Got it, you sort of channeled their experience. Have you ever shared a poem with someone about their own story? If so, how did they react?
No, I never did that, but I do open mics, so they have heard it, but they never know that it's their story. I tried to change a few things here and there so that they don't get to know that.
AI interviewer: So you keep it anonymous by changing details. If someone ever recognized their own story in your poem, how do you think you'd handle that conversation?
Transcript of the clip. The candidate's wording is kept as spoken; a place name is replaced with [city]. Interviewer lines in italics.
This is the typical competent candidate. They get everything across, but you can hear the effort throughout: restarts mid-sentence, plenty of fillers, simple connectors. None of it ever blocks the conversation.
Read the transcript (2:00)
I generally find ways to simplify the complex things in my work, like in today's AI world, we can use AI in preparing presentations, writing emails, all other stuff. So there are so many complex things, there may be some complex tasks or you need to repair a complex Excel sheet. So AI can use, we can use AI and we can present everything in a good and fun way. And by presenting very less efforts. So a fun fact about me is that I use the AI for the simplest of the simplest things as well as for the toughest of the toughest things. Just writing a one-line email also, I have been so much addicted with the AI, addicted with ChatGPT or Gemini that I sometimes use them because it saves me, it saves time, it saves basically for me to getting exhausted, it makes my productivity more better, but at times when there is no AI [unclear], no AI needed, I don't use it.
AI interviewer: So you mentioned you use AI tools even for simple things like one-line emails. Can you walk me through one specific example where using AI made a routine task surprisingly easier or more creative for you?
So in my current job, I need to write a lot of emails. I need to reply a various variety of emails and I need to resolve a queries of a lot of candidates. So what I do is like, I always, I have kept a few prompts ready with me and the AI now has understood what...
Transcript of the clip. The candidate's wording is kept as spoken; a place name is replaced with [city]. Interviewer lines in italics.
Workable, but you feel the effort. Simple topics go fine, while longer explanations start to strain, with persistent word searches, repeated fragments, and grammar that sometimes makes you reconstruct the sentence.
Read the transcript (2:00)
Okay, so basically I should say my story of the, like, the CP journey. So whenever, when I joined the college, so I know my college is like the tier two, tier three college. So I have to work hard to like, to like get into MNC or a startup, not because like my college don't get a well-deserved company and so on. So my plan was to like the build the profile so well, so even the [unclear] or fourth year, it's like a little or some founders see my profile, then they, they will, okay, the guy have some problems on the skills and they can do it. So I am, so I starting like the CP journey and I solve the Codeforces in my college years, like I really solve three, four, five questions a day, which is like the... at that time I think, oh, what I'm doing, like sometimes it's like [unclear] and hesitating me to do it. And sometimes it's fun to have a problem-solving skill. But at the end, but when I see now to back to my, like back journey, so it's like good to have fun. So like that I grab it and I can do it.
AI interviewer: All right, you mentioned solving problems daily was sometimes fun, sometimes tough. Can you tell me about one particular problem that really stuck with you? Maybe something memorable or a challenge you faced.
So yeah, so at the time, uh, this was the time at 2023. So I recently started the, uh, solving the Codeforces problem. And which, you know, the, like the Codeforces, like the, in the problems always skills, the Codeforces are super high, like the LeetCode was there and the CodeChef was there and I correct, but they are like the, from the majority of the, like the problems, uh... So like the, I'm not saying the India is not, but the global market and the global, like the question level are... level on Codeforces. So I do Codeforces. So at a time one...
Transcript of the clip. The candidate's wording is kept as spoken; a place name is replaced with [city]. Interviewer lines in italics.
Owner input needed before publishing (remove this box)
The samples above use the labelled clips from the communication analysis guide. Still needed from Fabric's own reports, not to be written by hand:
- For each sample: the exact score, the five dimension scores, two or three timestamped evidence notes, and the MTI read.
- Speaker context where approved (role applied for, channel). None is shown now.
- Consent: confirm each candidate agreed to public reuse with a transcript. Two transcripts carry personal details (a career history; a named city was replaced with [city]).
- Transcripts: check the [unclear] words against the audio.
- The accent pair below: two clips with different MTI reads and scores within 0.5 points.
Accent comparison: two different MTI reads
Owner input needed: two consented, anonymised clips with different MTI descriptions and scores within half a point of each other, with transcripts and timestamped evidence. The pair shows that accent description and English scoring are separate.
How to choose a spoken-English pass mark for your BPO roles
There is no single pass mark that fits every contact-centre role. A customer-facing voice process (hiring customer support executives, for example), a high-stakes relationship role, and an internal support role place different demands on spoken communication. Treat the figures below as starting points for calibration, not automatic hiring rules.
Find a starting pass mark for your role
Pick the role type. Each band says how to read a score there. Calibrate before you treat any line as a rule.
Customer-facing voice, support and sales
Escalations, key accounts and client-facing leads
Roles where most work happens in writing or inside the team
To calibrate a threshold, work through four steps with experienced trainers, using recordings from your own applicant pool.
Calibrate a pass mark in four steps
Avoid choosing a pass mark simply because it produces a convenient shortlist size. A threshold should reflect what agents need to do on the job. Keep English proficiency separate from accent, technical correctness, resume strength, and other selection criteria.
How this fits alongside Versant and SVAR
BPO hiring teams may already use Versant, SVAR, trainer-led assessments, or recruiter phone screens. Each approach has its own format and validation evidence. Versant, for example, publishes validation research comparing its machine scores with human ratings. That evidence deserves to be considered on its own terms; it should not be read as proof that another assessment produces equivalent scores.
Fabric’s practical proposition is a spoken-English assessment inside its phone-screening and AI-interview workflows. Both workflows use the same 0–10 scale, five dimensions, separate MTI read, and approximate CEFR reference. Every score cites the candidate’s words with timestamps, so a recruiter or trainer can inspect the evidence instead of relying only on an unexplained verdict. Re-scoring the same recording usually lands within about a third of a point, according to Fabric’s internal checks.
The right choice depends on your hiring workflow, the evidence you need to review, and how well the assessment aligns with your own trainers’ standards. Do not compare scores across tools as if they were interchangeable unless you have run a suitable same-candidate study. For a side-by-side view of the wider tool choice, see Fabric’s comparison of AI interview software for BPOs.
What this looks like at hiring-batch scale
When a team is hiring for a new batch, a consistent first assessment can help recruiters prioritise follow-up conversations and give trainers a shared language for reviewing communication. Fabric runs interviews around the clock, so assessment capacity does not have to be limited to office hours. Recruiters should still review the evidence and make decisions using role-related criteria. Integrity matters at batch scale too: in Fabric’s analysis of 19,368 AI interviews, 61% of cheaters scored above pass thresholds and would have advanced without detection.
If you are hiring voice-process agents, the most useful next step is to test the score against your own applicant pool. Review a sample of recordings with your trainers, agree on what good communication looks like for the role, and use that calibration to set your first pass mark. Fabric’s BPO and contact-centre hiring page shows the full workflow, from phone screening to the shortlist.
Frequently asked questions
1. What is a good spoken-English score for BPO hiring?
A score of 6+ is a suggested starting point for customer-facing sales or support roles, while some high-stakes relationship roles may start around 7.5+. These are starting points only; calibrate them with trainers and recordings from your own roles.
2. Is Fabric’s 0–10 score a CEFR score?
No. Fabric uses its own 0–10 scale, calibrated with CEFR descriptors and mapped approximately for reference. It is not a certified CEFR result.
3. Is the score equivalent to an IELTS band?
No. The IELTS figures shown are rough reference points, not an official conversion or IELTS-equivalent score.
4. Does an Indian or Filipino accent lower the score?
No. Accent itself is not scored. Fabric evaluates intelligibility and other spoken-English dimensions, while mother tongue influence is described separately.
5. What is mother tongue influence (MTI)?
MTI is a separate description of accent patterns such as sound substitutions and rhythm or stress carryover. It is descriptive, not a score, and does not change the English result.
6. Which dimensions does Fabric score?
Fabric scores intelligibility, fluency, vocabulary, grammar and coherence, each out of 10. The overall score is the average of the five dimension scores.
7. Does a candidate need to answer every question?
Skipping a question does not itself lower the English score. The assessment evaluates the speech available in the conversation; recruiters should separately decide whether the interview provides enough information to assess job fit.
8. Does thinking time affect the score?
Thinking time by itself does not affect the score. Pauses that indicate difficulty finding words may be relevant to fluency.
9. Can a longer answer get a higher score?
No. Answer length alone does not raise the score. The assessment focuses on the quality of spoken communication across the five dimensions.
10. What happens if the recording has background noise?
Background noise and audio quality are not English skills and should not lower the score. If the recording cannot be assessed reliably, obtain a clearer recording or use another assessment.
11. How consistent are scores when the same recording is rescored?
Fabric’s internal checks indicate that rescoring the same recording usually lands within about a third of a point. This is an internal consistency figure, not a claim that every score is exact.
12. Has Fabric been compared with human examiners?
In a comparison with the public Cambridge Speak & Improve learner dataset, rank correlation was 0.77 and 86% of speakers were within one CEFR level of an examiner. The dataset consists of learner recordings, not BPO calls, so the result should be interpreted with that limitation in mind.
13. Can I compare Fabric’s score directly with Versant or SVAR?
Not as if the scores were interchangeable. The tools use different assessment methods, and this article does not claim a same-candidate equivalence study. Compare the evidence, workflow fit and results against your own job requirements.
Sources and methodology
- Council of Europe: CEFR common reference levels, the level descriptors Fabric’s bands were calibrated against.
- Cambridge University Press & Assessment: Speak & Improve Corpus 2025, learner recordings annotated with CEFR levels by examiners; corpus paper by Knill et al. on arXiv.
- EF English Proficiency Index 2025, used as a cross-check on the band boundaries.
- Pearson: Versant by Pearson English Speaking and Listening Test, Test Description and Validation Summary (2024).
- Fabric’s communication analysis guide: the voice samples, band descriptions, production score distribution and rescoring consistency.
The production score distribution, the rescoring figure and the Speak & Improve comparison come from Fabric’s own evaluation. They have not been independently audited.