Krisp alternatives with on-premise deployment for a Fortune 500 in finance

Hey folks, hoping to tap into the collective wisdom here. I've been deep in the weeds evaluating Krisp for a large-scale, security-conscious environment, and while its noise cancellation tech is genuinely impressive for individuals, the cloud-centric SaaS model is a total non-starter for this particular use case. We're talking about a Fortune 500 financial institution where all voice/video data *must* remain on-premises due to regulatory and data sovereignty requirements.

So, I'm on the hunt for robust alternatives that offer Krisp-like background noise suppression (both mic and speaker) but can be deployed entirely within a private data center or VPC. The ideal solution would be something we can integrate into our existing virtual desktop infrastructure (VDI) and collaboration stack (think Webex, Teams rooms). Kubernetes-native or container-friendly is a huge plus for our platform engineering team.

Here's my preliminary research and some options I've been stress-testing in the lab:

**1. Open-Source / Framework Approach:**
* **RNNoise** (as a library): This is the backbone of a lot of noise suppression. We've done PoCs embedding it into custom-built audio pipelines. The control is fantastic, but it's heavy on the engineering lift.
```dockerfile
# Example Dockerfile snippet for a microservice wrapping RNNoise
FROM alpine:latest AS builder
RUN apk add build-base cmake git
RUN git clone https://github.com/xiph/rnnoise.git &&
cd rnnoise && ./configure && make install
# ... build our custom audio processor app
```
* **NVIDIA Maxine**: Their Audio Effects SDK provides noise removal and is deployable on-prem on NVIDIA hardware (GPU required). The quality is stellar, but you're locked into their hardware ecosystem.

**2. Commercial On-Premisable Solutions:**
* **Krisp for Enterprise?** – I've heard whispers of a possible on-prem version, but details are scarce. If anyone has concrete intel, I'm all ears.
* **Dolby Voice**: Offers an on-prem deployable media server (Dolby Voice Gateway) for conferencing. It's more of an end-to-end solution than a drop-in Krisp replacement, but the noise isolation is top-tier.
* **Agora SDK**: Provides real-time audio APIs with noise suppression that can be self-hosted. Their "Agora On-Premise Audio SDK" is worth a look, though the pricing gets complex.

**Key Requirements & Pitfalls I'm Watching:**
* **Latency:** Any added processing must be sub-50ms to not disrupt natural conversation flow.
* **CPU Overhead:** For VDI, we need something lightweight. Some libraries spike CPU on dense audio streams.
* **Integration:** Ideally, it's a system-wide audio filter (like Krisp) or a SDK that can be injected into our approved softphone and meeting apps.
* **Management:** Central policy control (who gets it, when) is a must.

Has anyone tackled a similar project for a regulated industry? I'm particularly curious about real-world performance in large-scale deployments (think 10k+ concurrent users) and how you handled the packaging/deployment—Helm charts, VM appliances, etc.

The trade-offs between building around an open-source core versus buying a supported commercial product are front of mind. The build route gives ultimate control but shifts the burden to our team for optimization and updates.

Any war stories, config snippets, or performance benchmarks would be incredibly helpful.

bw