This service takes the sound off a live video stream and delivers it as a continuous audio feed, prepared for whatever comes next — live transcription, translation, monitoring or recording.
Extraction itself needs no model and no GPU. What live work adds is the requirement never to fall behind: the audio must be pulled, cleaned and handed on continuously, with a fixed and small delay, and the pipeline must survive a stream that stalls, drops or changes format mid-flight. Cleaning is done with the lightest models available, because every millisecond spent on it is added to the delay before a word is transcribed.
Models in this group take a live camera or stream as input and must keep pace with it in real time. 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 how many camera streams or feeds one server of that tier can keep up with in real time, not a per-hour count: live work must fit inside the interval between frames, and a server that cannot keep pace drops frames rather than falling behind. 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 live 1080p stream with its audio extracted and prepared continuously.
FFmpeg
Vendor: FFmpeg project
What it does: pulls the stream, decodes it and writes a continuous audio feed at the required rate and channel layout, recovering from interruptions. No GPU needed, and it does the actual work.
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 (streams handled continuously)
≈ 40 streams
≈ 150 streams
≈ 400 streams
Silero VAD
Vendor: Silero
What it does: detects where speech is present so that downstream processing runs only when someone is talking, which multiplies the streams one server 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 (streams handled continuously)
≈ 60 streams
≈ 200 streams
≈ 600 streams
DeepFilterNet 3
Vendor: Friedrich-Alexander-Universität
What it does: removes steady background noise with negligible added delay, the usual live cleaning choice.
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 (streams handled continuously)
≈ 20 streams
≈ 70 streams
≈ 180 streams
RNNoise
Vendor: Xiph.Org Foundation
What it does: a classical suppressor needing no GPU, effective on hum and fan noise and free to run alongside everything else.
Requirement
Minimum
Medium
High
GPU type
No GPU required
No GPU required
No GPU required
VRAM
—
—
—
vCPUs
2
4
16
RAM
4 GB
8 GB
32 GB
Server
CPU instance, 2 vCPU
CPU instance, 4 vCPU
CPU instance, 16 vCPU
Rate (streams handled continuously)
≈ 100 streams
≈ 300 streams
≈ 900 streams
MDX-Net
Vendor: Kuielab
What it does: separates speech from music live where the stream carries a soundtrack, at a cost that limits how many streams one card can hold.
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 (streams handled continuously)
≈ 2 streams
≈ 8 streams
≈ 20 streams
Demucs v4
Vendor: Meta
What it does: the highest-quality separation, used live only where the stream must be split cleanly and the hardware for it is justified.
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 (streams handled continuously)
≈ 1 stream
≈ 4 streams
≈ 10 streams
Choosing between them
What is worth doing live depends on your delay budget and what the audio feeds. Our consultants review your streams and downstream task, then recommend the extraction settings, the cleaning that fits inside the delay allowed, and the handling for streams that misbehave.
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 to audio. 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.