// // PRODUCT — DUBSTREAM
Every language. The same second.
DubStream takes one live feed and returns dubbed audio in 40+ languages at 412ms glass-to-glass. Your audience picks a language and hears the speaker in it. Built, hosted and operated by ZebIQ.
//01
412ms
glass-to-glass, p95
//02
40+
languages live
//03
99.95%
uptime, rolling 30 days
//04
5,550
concurrent viewers, load-tested
Subtitles ask people to read. Dubbing lets them listen.
A summit with delegates from twelve states has one working language on stage and eight in the room. Subtitles help the people who read quickly in a second language. When a chief guest speaks for eighteen minutes without pausing, reading is work.
Human interpreters solve it properly and cost accordingly — booths, hardware, two interpreters per language per shift, and a logistics chain that has to arrive on time. Below a certain budget, most organisers simply drop the languages and accept that part of the room disengages.
DubStream sits between those options. One feed in, forty language tracks out, priced per language-hour so you only pay for the languages you actually run.
// WHAT DOES NOT CHANGE
- Your video. The original feed is never re-encoded through a model. Only the audio channel changes.
- Your player. The language selector embeds in your branded player or event app.
- Your production. We take a standard RTMP or SRT feed from the desk you already have.
- Your budget ceiling. A daily spend cap you set stops the engine rather than overrunning.
// PIPELINE
One feed in. Forty tracks out.
//01
Ingest
Your programme feed arrives over RTMP or SRT on redundant dual-path ingest. A 25-second socket idle-timeout means an encoder that loses internet reconnects cleanly instead of ending the stream.
//02
Transcribe
Dual-mode speech recognition runs an auto-detect stream alongside a fast fixed-language transcriber. The fixed path carries the latency budget; the auto-detect path catches language switches mid-sentence.
//03
Translate
Stabilised transcript segments fan out to one translation stream per target language. Segment boundaries are chosen so a sentence is never cut mid-clause.
//04
Synthesise
Each language gets its own neural voice track, time-aligned against the source so a listener never drifts out of step with the speaker on stage.
//05
Deliver
Listeners pick a language in a branded player or in your event app. Original video is untouched — only the audio channel changes.
// PRICING
Priced per language-hour. Nothing else.
No platform fee, no per-seat licence, no minimum contract. You pay for the hours you run, in the languages you choose.
// STANDARD
₹8,000/ language-hour
Single events and short runs. Everything included — rehearsal, live operations desk, branded selector, recordings.
// VOLUME
from₹5,000/ language-hour
Multi-day events, multi-city series and annual programmes. Rates scale with committed language-hours across the year.
// WORKED EXAMPLE
A one-day summit
8 programme hours × 6 languages = 48 language-hours.
₹3,84,000
at standard rate
A three-day conference
21 programme hours × 4 languages = 84 language-hours.
₹5,46,000
at ₹6,500, volume tier
An annual programme
Committed hours across a year of events reach the floor rate.
₹5,000
per language-hour
Figures are indicative and exclude GST. Final pricing depends on committed hours, languages and whether you need on-site engineering alongside the platform.
// INCLUDED AT EVERY TIER
The rate is the whole price.
Rehearsal and dry run
We test your actual venue uplink and encoder before show day, not a lab equivalent. Failover is drilled, not assumed.
Operations desk during the event
An engineer watches latency, drift and per-language quality live, with a documented escalation path.
Branded language selector
Embedded in your player, your event app or a standalone link. Your branding, not ours.
Recordings per language
Every dubbed track is retained and handed over after the event for VOD, social clips or compliance.
Glossary and name handling
Speaker names, product names and technical terms are loaded ahead of time so they are not mangled live.
Daily spend cap
A hard ceiling you set. The engine stops rather than quietly running past your budget.
// QUESTIONS
The things people ask before they book.
How is a language-hour counted?
One hour of one language being actively dubbed. A six-hour session in four languages is 24 language-hours. We bill on scheduled programme time, not wall-clock time — breaks and lunch are not charged unless you want the channel live through them.
Is 412ms actually noticeable in the room?
It is close enough that a delegate listening on headphones inside the hall stays in step with the speaker on stage. It is not zero. For events where absolute lip-sync matters more than immediacy, we can add a small fixed delay to the video so the two align exactly.
What happens if the internet at the venue drops?
Ingest runs on dual paths with automatic failover, and the socket timeout is tuned so an encoder that briefly loses connectivity reconnects rather than ending the session. If both paths fail, the original stream continues without dubbing and the language tracks resume when connectivity returns. We do not pretend an uplink cannot fail; we plan for it.
Can it handle a speaker who switches between English and Hindi mid-sentence?
Yes, and this is common at Indian events. The dual-mode recogniser runs an auto-detect stream specifically for this. Accuracy on rapid code-switching is good but not perfect, and we will say so in the rehearsal rather than after the event.
Do you replace human interpreters?
For most corporate and government events, yes, at a fraction of the cost and with no booth or hardware logistics. For legal proceedings, medical consent, or anything where a mistranslation carries liability, use a human interpreter. We will tell you when that is the right call.
Also see multi-language event support and live streaming infrastructure.
Run your next event in every language your audience speaks.
Send us the programme length and the languages. We will come back with a fixed number and a rehearsal date.