# FreunDo

> FreunDo develops low-latency, low-power FPGA-accelerated edge AI and lightweight speech enhancement technology for hearing aids, assistive listening devices, and intelligent sensing systems.

Website: https://freundo.co.uk/
Contact: contact@freundo.co.uk
Location: Edinburgh, United Kingdom

## Priority Pages

- Homepage: https://freundo.co.uk/
- Machine-readable company content: https://freundo.co.uk/content.md
- Services and capabilities: https://freundo.co.uk/services.md
- Commercial engagement and pricing model: https://freundo.co.uk/pricing.md
- AI capability manifest: https://freundo.co.uk/ai-capabilities.json
- Sitemap: https://freundo.co.uk/sitemap.xml
- Robots policy: https://freundo.co.uk/robots.txt

## What FreunDo Builds

FreunDo builds model-to-hardware edge AI acceleration for compact devices. The first focus area is real-time speech enhancement for hearing aids and assistive listening devices.

Core technology:

- RFD lightweight speech denoising and enhancement model
- RepConv reparameterisation for lean deployment
- Causal attention for low-latency streaming inference
- Frequency-heterogeneous output head for separate low-band and high-band handling
- FPGA reference designs, SDK support, and integration services

## RFD Model Facts

- Input: noisy speech
- Output: enhanced speech in real time
- Sample rate: 16 kHz
- Hop: 16 ms
- Compute: 28.3 MMAC/s
- Parameters: 29k during training, 23.7k deployed
- Latency design: causal streaming with no look-ahead
- VCTK-DEMAND test set: 824 utterances at 16 kHz
- PESQ: 2.30
- STOI: 0.925
- SI-SNR: 15.17 dB

## Commercial Fit

FreunDo is relevant for:

- Hearing aid OEMs
- Assistive listening device companies
- Audio SoC and semiconductor teams
- ODM suppliers
- Embedded sensing and wearable device companies

Engagement options:

- Technical demo
- Paid proof of concept
- FPGA reference design evaluation
- Non-recurring engineering support
- IP licensing and integration support