Model reference — Video · Live feed
This service transcribes the speech in a video and returns text with timings, so that footage becomes searchable and quotable — recorded meetings, lectures, broadcasts, body-worn and interview recordings, and published media.
The pipeline is extract, optionally clean, transcribe, and align. Alignment is what makes the result useful on video specifically: each word carries the moment it was spoken, so a search leads straight to the frame. Where several people speak, the transcript is also divided by speaker. Whether transcription runs after the fact on files or live as the words are spoken is the decision that shapes everything else, because live work sets a latency budget that rules out the slower and more accurate models.
Models in this group take a recorded video file as input and are applied frame by frame. Each specification table gives three hardware tiers — Minimum, the smallest setup on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is the number of sample inputs processed per hour on that hardware. Use these rates for initial sizing. Before production, benchmark your own data to validate accuracy, latency, throughput and cost. The sample input here is one minute of 1080p video at 25 frames per second.
Vendor: SYSTRAN
What it does: transcribes accurately in roughly a hundred languages, re-engineered to run several times faster than the original with identical output. The usual default for recorded video.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 1,200 | ≈ 4,200 | ≈ 10,800 |
Vendor: OpenAI
What it does: the reference model, slightly more accurate on difficult audio at roughly a fifth of the speed. Used where a missed word matters more than cost.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 240 | ≈ 720 | ≈ 1,800 |
Vendor: NVIDIA
What it does: transcribes and translates major European languages in one model, giving both the original words and an English version from a single pass.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 900 | ≈ 3,000 | ≈ 7,200 |
Vendor: University of Oxford (VGG)
What it does: aligns the transcript to the audio word by word, which is what lets every line link to the exact frame and makes subtitles time correctly.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 900 | ≈ 3,300 | ≈ 8,400 |
Vendor: pyannote (Hervé Bredin)
What it does: divides the transcript by speaker so a multi-person recording reads as a conversation with attributions.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 6 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 1,200 | ≈ 4,200 | ≈ 10,800 |
Vendor: FFmpeg project
What it does: extracts and prepares the audio track at the rate the transcription model expects. No GPU needed, and the step that most often silently costs accuracy when misconfigured.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | No GPU required | No GPU required | GPU-accelerated decode (NVENC/NVDEC) |
| VRAM | — | — | 8 GB |
| vCPUs | 2 | 8 | 16 |
| RAM | 4 GB | 16 GB | 32 GB |
| Server | CPU instance, 2 vCPU | CPU instance, 8 vCPU | 1× RTX 4090 24 GB |
| Rate (video minutes processed per hour) | ≈ 6,000 | ≈ 20,000 | ≈ 60,000 |
Models in this group take a live camera or stream as input and must keep pace with it in real time. The three hardware tiers mean the same as above, but the rate here is how many streams one server of that tier can keep up with in real time rather than a per-hour count. The sample input here is one 1080p stream at 25 frames per second.
Vendor: NVIDIA
What it does: transcribes English in real time with very low latency, which is what live captioning requires. The choice for live streams.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3090 | RTX 4090 | A100 80 GB |
| VRAM | 12 GB | 24 GB | 80 GB |
| vCPUs | 8 | 12 | 24 |
| RAM | 32 GB | 48 GB | 96 GB |
| Server | 1× RTX 3090 24 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 8 streams | ≈ 30 streams | ≈ 80 streams |
Vendor: SYSTRAN
What it does: transcribes live in a hundred languages using short overlapping windows, at a latency of a few seconds — acceptable for monitoring, too slow for on-screen captions.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 3 streams | ≈ 12 streams | ≈ 32 streams |
Vendor: Silero
What it does: detects speech so transcription runs only when someone is talking, which multiplies the number of streams one card can carry.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 4 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 8 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 60 streams | ≈ 200 streams | ≈ 600 streams |
Vendor: Friedrich-Alexander-Universität
What it does: cleans the live audio before transcription with negligible added delay, which is usually the cheapest accuracy improvement available live.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 4 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 8 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (camera streams handled at 25 frames per second) | ≈ 20 streams | ≈ 70 streams | ≈ 180 streams |
Model choice depends on your audio quality, languages and whether transcription must be immediate. Our consultants measure word error rate on your own footage, then recommend a model, the preparation worth doing and the hardware for your volume.
At the start of a project we may run a short proof of concept on a sample of your own data, measuring the accuracy and the throughput the model actually achieves on your material. That replaces the estimates on this page with real figures, so the cost and the schedule for the full engagement are known before it is committed.
Share a representative sample, expected volume, latency target and deployment location for Video audio to text. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster.