Model reference — Audio
Audio classification labels what a recording contains — speech, music, an alarm, breaking glass, a machine running rough, a particular species of bird — and tagging attaches several such labels with the times at which each occurs.
This is how large volumes of audio become searchable and how sound becomes a monitoring signal. A factory recording can be checked for the sound a failing bearing makes; a site recording for an alarm nobody logged; a media archive for every stretch containing music that must be cleared for rights. Off-the-shelf models cover several hundred common sound categories. Where the sound you care about is specific to your equipment, a small model is fitted to examples of it, which needs recordings of both the normal and the abnormal sound.
Models in this group take an audio file as input: a recording, a call, an interview or a broadcast. 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 hour of 16 kHz mono speech.
Vendor: Microsoft
What it does: labels audio across several hundred sound categories with the best general accuracy available. The default for tagging a media or monitoring archive.
| 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 (audio hours processed per hour) | ≈ 60 | ≈ 220 | ≈ 550 |
Vendor: Massachusetts Institute of Technology
What it does: a well-established classifier over the same broad category set, easy to fit to your own categories from a few hundred examples.
| 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 (audio hours processed per hour) | ≈ 80 | ≈ 300 | ≈ 800 |
Vendor: University of Surrey
What it does: a family of pretrained classifiers widely used as a starting point for custom sound detection, well documented and dependable.
| 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 (audio hours processed per hour) | ≈ 90 | ≈ 340 | ≈ 900 |
Vendor: Google
What it does: a very small classifier covering 521 everyday sound classes, light enough to run continuously on modest hardware at a remote site.
| 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 (audio hours processed per hour) | ≈ 400 | ≈ 1,400 | ≈ 4,000 |
Vendor: Microsoft and LAION
What it does: matches audio against a written description — "a metallic scraping sound", "a smoke alarm" — with no training examples at all, which lets a new sound be searched for the same day it is described.
| 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 (audio hours processed per hour) | ≈ 70 | ≈ 260 | ≈ 700 |
Vendor: Silero
What it does: detects where speech is present, which is run first so that classification and tagging are applied only to the stretches that matter.
| 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 (audio hours processed per hour) | ≈ 600 | ≈ 2,000 | ≈ 6,000 |
The choice depends on whether your categories are common sounds or specific to your equipment, and whether detection must be immediate. Our consultants review your recordings and the events you need flagged, then recommend either an off-the-shelf model or a fitted one, with the alerting threshold to go with it.
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 Audio classification / tagging. 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.