Chatterbox

Live voice translation for customer calls

Speak your language. Be understood in theirs. Real-time, two-way voice translation for contact center agents.

00:00 / 00:00

What it does

Real-time voice translation for one-to-one conversations

Both sides speak their own language, inside the live call.

Speech to speechBoth directions

Availability

Desktop app for contact center agents

Installs on the agent's machine and works with the tools you already use.

Desktop app
0

Languages

0.00s

Median speech-onset latency

0.00s

Time to finalized text

Why Chatterbox

Language should never decide who you can serve

Multilingual customer support shouldn't depend on who is on shift. Hiring native speakers or adding interpreters adds cost and delay. Chatterbox removes that barrier: customers hear a fluent conversation, not a machine relay.

Use cases

Live voice translation where your conversations happen

Contact center calls

Agent in English, customer in German: each hears their own language. Works with CPaaS platforms.

One-to-one online meetings

Select Chatterbox Virtual Mic in Zoom, Microsoft Teams or Google Meet. Each side hears the other in their own language.

What sets it apart

Not a relay. A real conversation.

01 / 04

Tone, intent and nuance survive the translation

Speech-to-speech translation designed to preserve tone, intent and conversational nuance, so an apology still sounds like an apology.

02 / 04

No interpreter in the middle

No scripted turns, no third person on the line, no waiting for a relay.

03 / 04

A live transcript on the agent's screen

A real-time written translation of both sides runs alongside the call.

04 / 04

Validated in real operations

Proven through an extensive 12-month technology evaluation, including live-environment validation and large-scale contact center pilots.

WHAT SETS IT APART
Agent · EN-USTone: apologetic

I'm so sorry for the wait. Let me fix this for you right now.

Customer hears · IT-ITTone: apologetic

Mi scuso davvero per l'attesa. Lo sistemo subito per lei.

Live callBoth directions · Live
Agent · EN
Customer · IT
live
00:0000:0400:08
VOIS pilot: 10,000 agents after a 12-month evaluation

Features

Built on our own speech and voice models

Chatterbox is a multilingual voice platform powered by our own BOLI and MARS models, built for superior performance at low latency.

  • 60 languages

  • Business-specific glossaries

  • Voice cloning

  • Regional accents

  • Multiple deployment options

  • API support

Benchmark results

First to be heard

Six real-time translation systems, side by side. Chatterbox starts speaking first.

Chatterbox leads median speech-onset latency

Responsiveness that shows up in everyday interactions. Chatterbox leads on both median and average speech-onset time, delivering a strong foundation for responsive multilingual conversations.

50% of tested requests start producing speech within this time. Represents the typical user experience.

Seconds, lower is better.
Chatterbox
3.74s
Gemini
4.25s
Soniox
5.93s
Palabra
6.71s
Speechmatics
7.85s
xAI
⫽14.07s

Seconds, lower is better.

Chatterbox
3.74s
Gemini
4.25s
Soniox
5.93s
Palabra
6.71s
Speechmatics
7.85s
xAI
⫽14.07s

Chatterbox leads median speech-onset latency

Responsiveness that shows up in everyday interactions. Chatterbox leads on both median and average speech-onset time, delivering a strong foundation for responsive multilingual conversations.

50% of tested requests start producing speech within this time. Represents the typical user experience.

Seconds, lower is better.

Chatterbox
4.27s
Gemini
4.44s
Soniox
6.17s
Palabra
7.10s
Speechmatics
8.70s
xAI
⫽19.47s

Chatterbox has the lowest average speech-onset latency

Less waiting. More conversation. Chatterbox starts translated speech sooner on average than every other system tested, helping customers and agents maintain conversational momentum.

Average time from the defined request/input trigger until the system starts producing translated audio output across all test samples.

Seconds, lower is better.

Gemini
5.76s
Chatterbox
6.92s
Soniox
9.20s
Palabra
11.21s
Speechmatics
15.57s
xAI
⫽50.92s

Chatterbox ranks second on P95 speech-onset latency

Strong responsiveness beyond the typical interaction. Chatterbox combines leading average response speed with a P95 latency lower than four of the five competing systems tested.

95% of requests start producing speech within this time; only 5% take longer. Useful for measuring production consistency.

Seconds, lower is better.

Chatterbox
2.67s
xAI
2.82s
Gemini
3.91s
Soniox
4.37s
Palabra
6.49s
Speechmatics
7.48s

Chatterbox delivers the fastest finalized text

Get complete text sooner so the next step can begin. Chatterbox delivers finalized text faster than every other system tested, helping downstream tools access the information they need earlier.

Time until the system returns the first finalized/stable transcription or translation result that is no longer expected to change.

Seconds, lower is better.

Speechmatics
2.30s
Soniox
2.34s
Chatterbox
2.67s
Palabra
2.75s
xAI
2.82s
Gemini
3.91s

Chatterbox ranks third on first partial result

Time from the start of speech/audio processing until the system returns its first partial/interim transcription or translation result. Measures responsiveness.

Measure (seconds, lower is better)ChatterboxGeminiSonioxPalabraSpeechmaticsxAI
Median speech-onset latency (TTFS P50)50% of tested requests start producing speech within this time. Represents the typical user experience.3.744.255.936.717.8514.07
Average speech-onset latency (TTFS avg)Average time from the defined request/input trigger until the system starts producing translated audio output across all test samples.4.274.446.177.108.7019.47
P95 speech-onset latency (TTFS P95)95% of requests start producing speech within this time; only 5% take longer. Useful for measuring production consistency.6.925.769.2011.2115.5750.92
Finalized text (TTFfinal)Time until the system returns the first finalized/stable transcription or translation result that is no longer expected to change.2.673.914.376.497.482.82
First partial result (TTFpartial)Time from the start of speech/audio processing until the system returns its first partial/interim transcription or translation result. Measures responsiveness.2.673.912.342.752.302.82

Case study

Voice without borders

Contact center agent wearing a headset during a live customer callVOIS

Piloting real-time multilingual support with VOIS

VOIS is piloting real-time multilingual support in select European contact centers, with Chatterbox performing real-time speech-to-speech translation.

0K
Agents in the VOIS EU pilot
0 mo
Technology evaluation before the pilot
0 + EN
German, Czech and Ukrainian alongside English
Read the case study

60 languages

Talk to your customers in their own language

Any supported language can pair with any other. Search for yours below.

EnglishItalian

Integrations and deployment

Plugs in. Deploys your way.

Integration

Select "Chatterbox Virtual Mic" as your input

Google Meet Zoom Microsoft Teams CPaaS platforms

Deployment

Three deployment models, your choice

Public cloudPrivate cloudOn-premises

Chatterbox runs as a desktop app on the agent's computer.

Getting started

How to translate a call in four steps

FAQs

Frequently asked questions

Live voice translation turns speech into another language while the conversation is happening. With Chatterbox, each person speaks their own language and hears the other in theirs, with a live transcript on the agent's screen.

No. Chatterbox runs on the agent's side only. The customer joins the call normally and hears a translated voice in their own language.

Yes. Chatterbox registers as a virtual microphone, so you select Chatterbox Virtual Mic in Zoom, Microsoft Teams or Google Meet. It also works with CPaaS platforms.

Not today. Chatterbox is built for one-to-one conversations, which is the shape of customer care calls. To translate a live stream for a whole audience, use live dubbing.

60 languages, and any supported language can pair with any other. New languages will be consistently added.

In a benchmark of six tested systems, Chatterbox leads on both median and average speech-onset time: 3.74 seconds median and 4.27 seconds average.

Chatterbox is built for live business calls. It runs as a desktop app on the agent's computer and works as a virtual microphone inside Zoom, Microsoft Teams, Google Meet and CPaaS platforms. It translates both sides out loud in real time, shows the agent a live transcript, and supports business-specific glossaries and voice cloning.

Yes. Chatterbox is available in public cloud, private cloud and on-premises deployments.