Real-Time Voice Translation for Customer Support | Edith

Real-time voice translation for customer support: how Edith runs live multilingual calls

By Arsalan Kamran, CEO and co-founder of AssistRing. Last updated: July 2026.

Edith is AssistRing’s real-time voice translator. It sits inside a live customer call and converts speech in both directions as people talk, so an agent and a customer who do not share a language can still have a normal conversation. No interpreter on the line. No “please hold” while someone translates.

We did not build Edith to demo it at conferences. We built it because we were losing good conversations over a language gap, and the tools we could buy were either too slow to use on a live call or could not handle the mix of languages our clients needed. This is what Edith does, how it works on a real call, and how it compares to the usual ways support teams try to cover more than one language.

What is real-time voice translation?

Real-time voice translation is software that listens to spoken words in one language and speaks them back in another while a call is happening. Unlike subtitles or a text chatbot, it works on live audio in both directions, so both people hear the conversation in their own language with only a short delay.

That last part matters. A lot of tools translate only one side, or work only in chat, or need a human interpreter dialled into the call. Edith is built for the harder version of the problem: two people talking on a live phone call, each hearing the other in their own language, without a third person slowing everything down.

Why language barriers cost support teams money

This is not a soft, feel-good problem. It shows up in revenue.

In a survey of 8,709 consumers across 29 countries, CSA Research found that 76% of shoppers prefer to buy products with information in their own language, and 40% will never buy from a business that only offers another language. Three in four say they are more likely to buy from the same brand again when customer care is handled in their language. CSA Research (formerly Common Sense Advisory) has also reported that 64% of executives lost online sales because of misunderstandings caused by a language barrier.

We saw the same thing from the operations side before we saw the research. Calls that should have taken five minutes ran three times longer. Agents got frustrated and checked out. Client escalations that looked like service failures were really translation gaps. Half the languages our QA reviewers were meant to score, they could not actually read, which meant quality was basically ungraded in those languages. Edith was our answer to the live-call half of that problem.

How Edith works on a live call

The point of Edith is that an agent can take a call in a language they do not speak and still sound human. Here is what happens in practice:

1. The agent sets the customer’s language, or Edith detects it.
   
   
2. The customer speaks. The agent hears it in English.
   
   
3. The agent replies in English. The customer hears it in their language.
   
   
4. Both directions run through the same live call, so the back-and-forth keeps its natural rhythm.
   
   

Edith is built to hold professional-level fluency across major languages, not the rough, word-salad output people expect from free translation tools. The goal was never “good enough to get by”. It was a call the customer would not realise had been translated at all. That is a high bar, and it is the bar we work to.

Edith compared to the usual ways teams cover multilingual support

Most support teams reach for one of three options when a new language shows up. Each one works to a point. Here is how they stack up against a live translator like Edith.

Approach	Time to add a language	Cost to scale	Works live on the call	Conversation flow
Hire native-speaker agents	Weeks to months per language	High. One hire per language, per timezone	Yes	Natural
Phone interpreter service	Fast to start	High. Per-minute billing adds up quickly	Yes, but a third person is on the line	Slow and stop-start
Generic machine translation	Fast	Low	Usually text only, one direction	Robotic, easy to misread
Edith (AssistRing)	Immediate for supported languages	Flat. One team covers many languages	Yes, both directions, no third party	Natural, kept close to human pace

The honest read: hiring native speakers is still the gold standard for a single high-volume language, and we will tell you when that is the right call. Edith wins when you need breadth, when you need it now, or when the economics of hiring a separate specialist for every language stop making sense.

Who Edith is for

Edith earns its keep anywhere a support team meets customers in more than one language:

• • E-commerce brands selling across Europe and North America, where a single store can pull in Spanish, German, French, and Dutch speakers in the same week.

• • Hospitality teams handling reservations and guest issues from international travellers.

• • Healthcare and patient intake, where getting the details right in the patient’s own language is not optional.

• • SaaS support teams whose users are global but whose support desk was only ever staffed in English.
    
    

Built by a BPO, for a BPO

Plenty of translation software is built by engineers who have never run a support floor. Edith was not. It came out of our own delivery operation across the US, Germany, and Pakistan, where the language problem was a daily reality rather than a slide in a pitch deck.

Edith also has a sibling. Sage, our AI QA evaluator, scores calls consistently across languages, so quality stops being a blind spot in every language except English. Together they cover the two places where multilingual support usually falls apart: the live conversation and knowing whether that conversation was any good. Both were built to fix our own problems first, which is the main reason we trust them on client work.