Real-time AI speech-to-text captioning.
Mandarin, Taiwanese, Hakka and English, live.
Point Media Tech's STT is an AI speech-to-text captioning engine built for broadcast, recognising mixed Mandarin, Taiwanese, Hakka and English in real time. End-to-end latency under 500ms; caption accuracy up to 98%. Live news, sports and post-production share one captioning pipeline, with output formats compatible with Taiwan NCC caption rules.
< 500 msEnd-to-end latency
98%Caption accuracy
4Languages mixed
24/7Unattended captioning
It keeps up with the line
where the guest switches language.
Every STT transcript line carries a timecode, a speaker and a language tag — including the one with an English term inside a Mandarin sentence. The glossary panel shows the corrections being forced in real time: that is the gap between 91% and 99%.
From manual stenography and post-edit cleanup
to AI real-time captioning, captions become part of transmission.
STT moves captioning out of post and into the transmission chain: live captions in real time, VOD caption files in batch. The legacy workflow staffs two stenographers per live broadcast and outsources VOD post-edit cleanup, so a 30-minute news segment takes four hours to caption while budgets shrink and stenographers get harder to hire. STT runs inference on GPUs in under 500ms, with mixed-language recognition (Mandarin + Taiwanese + Hakka + English), automatic punctuation, and SRT / EBU-STL output that drops straight into Playout / MAM.
Stenographers, post-edit cleanup, outsourced cost
Two to three stenographers per live broadcast; VOD captioning outsourced; a 30-minute show takes four hours of cleanup; Taiwanese and Hakka cost extra; nights and weekends bill at premium; staff turnover breaks terminology continuity.
Real-time, multilingual, automated output
GPU-based real-time inference under 500ms; mixed Mandarin / Taiwanese / Hakka / English recognition; automatic punctuation and segmentation; custom terminology and accent models; SRT / VTT / EBU-STL output drops straight into Playout / MAM; 24/7 unattended.
Today · Typed by hand, proofread afterwards
- Two stenographers on shift for a live programme
- VOD subtitles outsourced for proofing
- Four hours to subtitle a 30-minute programme
- Budgets tighten while the people get harder to find
After · Live AI subtitling
- Recognition in real time, under 500 ms
- Mandarin, Taiwanese, Hakka and English mixed freely
- Punctuation and segmentation applied automatically
- SRT / EBU-STL delivered straight to playout and MAM
Three layers, one continuous
real-time captioning pipeline.
STT works in three layers: multilingual recognition, real-time inference and broadcast integration. Each layer is sized independently, licensed by channel and language model, and extensible as you grow.
Multilingual ASR · Multilingual ASR
Mandarin (traditional / simplified), Taiwanese (Hokkien), Hakka, English, Cantonese with mixed recognition; automatic language switching; custom broadcast terminology training; news / sports / variety / drama acoustic models; smart transcription of numbers, proper nouns, and acronyms.
- Mandarin / Taiwanese / Hakka / English / Cantonese
- Automatic language switching
- Custom terminology + scene models
- Accent + background-noise robustness
Real-time inference · Real-time Inference
GPU-accelerated inference with end-to-end latency under 500ms; automatic punctuation and segmentation; speaker diarization; multi-channel concurrency; node failover; 24/7 uninterrupted service.
- < 500ms end-to-end latency
- 98% caption accuracy
- Speaker diarization
- Multi-channel + automatic failover
Broadcast output · Broadcast Output
SRT / VTT / TTML / EBU-STL output; direct delivery to Playout systems and Marquee Graphics; write-back to MAM as time-coded indexes; REST API and webhooks; NCC caption rule compatibility; HLS / DASH OTT platform support.
- SRT / VTT / TTML / EBU-STL
- Playout / Marquee / MAM integration
- REST API + webhooks
- NCC caption rule compatible
Speech recognition
Punctuation and segmentation
Subtitle formatting
Push to playout / OTT
End to end under 500 ms
The STT captioning flow: audio in, captions out,
five steps, fully automated.
STT takes a live signal or VOD file, runs GPU speech recognition, punctuation and formatting, and delivers captions to Playout, Marquee Graphics, MAM and OTT platforms.
- Audio InLive signal / VOD file · multi-channel
- ASRGPU inference · multilingual mixed recognition
- PunctuatePunctuation + segmentation + diarization
- FormatSRT / VTT / EBU-STL · broadcast-spec
- DeliverPlayout · Marquee · MAM · OTT
Engineering specifications.
STT deploys in containers on NVIDIA GPUs, and channel count scales with the GPU. A full spec sheet and live engineering walkthrough are available on request.
| Languages supported | Mandarin (traditional / simplified) · Taiwanese · Hakka · English · Cantonese · automatic switching · custom terminology |
|---|---|
| Recognition performance | End-to-end latency < 500ms · caption accuracy 98% (standard scenarios) · speaker diarization |
| Input formats | SDI / NDI / RTMP / SRT live streams · WAV / MP4 / MXF VOD files · multi-channel concurrency |
| Output formats | SRT · VTT · TTML · EBU-STL · CEA-608/708 · live RTMP / SRT caption tracks |
| System integration | Playout · Marquee Graphics · MAM · REST API · webhooks · OTT (HLS / DASH) |
| Hardware requirements | NVIDIA GPU (T4 / A10 / L4 entry-level) · Linux · containerized deployment · channel count scales with GPU |
| Licensing | Licensed per speech recognition channel and language model; monthly or annual subscription; software license + annual maintenance |
Three steps, from evaluation to production.
Request a trial
Submit the contact form. Sales will respond within one business day to schedule a technical consultation.
Integration assessment
Our consultant reviews channel count, language model needs, and integration with your existing Playout / Marquee / MAM to size the GPU configuration and trial plan.
Deploy and go live
After GPU deployment and terminology customization, we run acceptance tests covering both live and VOD scenarios, then phase the rollout to keep your captioning workflow running.
Frequently asked questions.
Which languages does STT support, and can it handle mixed Mandarin and Taiwanese?
Yes. Point Media Tech's STT supports Mandarin (traditional / simplified), Taiwanese (Hokkien), Hakka, English and Cantonese, switching language automatically within a sentence. Terminology and accent models can be customised for news, sports, variety and drama to reduce the impact of background noise and overlapping speakers.
How does AI real-time captioning compare with manual transcription?
STT generates captions in real time and can run unattended 24/7, so live broadcasts no longer need a stenographer rota. End-to-end latency is under 500ms, caption accuracy is 98% in standard scenarios, and it is more than 10x faster than manual transcription. It suits live news and sports, and batch caption generation in post-production.
Which caption formats does STT output, and does it meet NCC caption rules?
STT outputs SRT, VTT, TTML, EBU-STL and CEA-608/708 captions in formats compatible with Taiwan NCC caption rules. It also carries live caption tracks in RTMP / SRT streams and delivers captions straight to playout systems, media asset managers (MAM) and HLS / DASH OTT platforms.
How is STT licensed?
STT is licensed per speech recognition channel and language model, on monthly or annual plans with annual maintenance. Contact sales for detailed pricing matched to your captioning workload.
What hardware does STT need?
STT runs real-time AI inference on NVIDIA GPUs (T4 / A10 / L4 and up), on Linux in containers. Concurrent channel count scales with the GPU, and our technical consultants will help size the configuration.
How do I start evaluating and trialing STT?
Submit the contact form and sales will respond within one business day to schedule a technical consultation on channel count and language model needs. Rollout starts with terminology customisation, followed by acceptance tests on live and VOD scenarios.
Ready to make captions
follow the voice?
Book a demo and see STT captioning live news, sports and post-production in real time.