greater average SCOREQ improvement than Adobe Podcast
Diffio01 · Investment thesis · August 2026
Diffio turns damaged speech into usable media for creators and the products they rely on.
One-click generative restoration for noisy, echoey, distorted, clipped, and bandwidth-starved recordings.
company-reported ARR as of August 2026
Production web app + REST API · Bootstrapped · $0 outside funding
02 · Customer problem
Bad capture turns publishing into rescue work.
Creators, production teams, and media-tool builders lose time to filter chains, re-records, and audio that still cannot ship.
- 01DiagnoseNoise, echo, clipping, distortion, bandwidth loss
- 02PatchStack tools, tune controls, render, listen, repeat
- 03CompromiseSpend expert time or publish speech people abandon
03 · Product transformation
Upload once. Compare the real result.
Diffio restores speech in the web app or through the production API. Switch between the original and restored Amelia Earhart broadcast below.
Real published Diffio demo · Use headphones for the clearest comparison
Loading waveform...
Loading audio enhancer...
04 · Reproducible proof
Diffio wins across quality and intelligibility—not one cherry-picked clip.
A public, reproducible benchmark assembled by Diffio tested 100 archival clips at 30 seconds each.
Adobe Podcast
1.7221
Diffio 3.5
2.1102
+22.54% average improvement
- SCOREQ wins
- 83/100
- Mean proxy WER
- 0.1253 vs 0.2079
- WER wins
- 79/100 + 14 ties
- DNSMOS OVR wins
- 69/100
SCOREQ estimates mean opinion score; WER is a frozen-ASR intelligibility proxy. Results are model-based, not a human listening panel. Methodology and data ↗
05 · Traction
Revenue arrived before outside capital.
A solo founder shipped the model, product, billing, and production API—and converted customers to recurring plans.
Company-reported · August 2026≈ $9,000 ARRBootstrapped · $0 outside funding
- Paying Basic subscribers
- Consumer web app in production
- REST API + Python and Node SDKs
- One employee · founded 2024
06 · Why now
More speech is published—and more valuable archives are waiting.
Demand signals, not a top-down TAM.
167MAmericans consume podcasts monthlyEdison, 2026 ↗812MU.S. podcast hours each weekEdison, 2026 ↗$44Bprojected U.S. creator ad spendIAB, 2026 estimate ↗3M+recorded-sound items at the Library of Congress+75k yearly ↗
WedgeHard recordings for independent creators→ExpandProduction teams and back catalogs→PlatformRestoration inside media products
07 · Business model + GTM
Self-serve proves value; usage expands with the customer.
60 audio minutes / month
6 → 20 hours / month, audio or video
$5 starter credits · $5/hr Diffio 2.0 · $16/hr Diffio 3.5
Initial buyer: independent podcast/video producers with difficult speech.
Expansion: teams add volume; developers embed restoration through keys, SDKs, webhooks, spend controls, and usage tooling.
Strategy: turn benchmark-led discovery and product proof into repeatable creator acquisition and API design partnerships.
App subscription dollar prices load live from Stripe and are intentionally not duplicated here.
08 · Competition
Diffio is the restoration specialist—not another editing suite.
| Alternative | Strong fit | Diffio distinction |
|---|---|---|
| Adobe Podcast | Fast browser cleanup inside Adobe’s creator ecosystem. | Public hard-audio benchmark, model choice, and self-service API. |
| Descript Studio Sound | Transcript-first podcast and video editing. | Focused restoration without requiring an editing suite. |
| Auphonic | Loudness standards, leveling, and broadcast workflows. | Generative reconstruction for more severely degraded speech. |
| Cleanvoice | Filler-word, silence, and podcast edit automation. | Restoration quality is the primary product, with a self-service API. |
| Manual restoration | Expert control for bespoke, high-value material. | One-click throughput and consistent web/API workflow. |
| Do nothing / re-record | Acceptable when quality does not matter or capture can be repeated. | Recovers value when the original moment cannot be recreated. |
No single tool wins every workflow. Diffio focuses on difficult speech, self-service access, and web-to-API continuity.
09 · Technical moat
The core model is proprietary; the evidence is inspectable.
Diffio is not a wrapper around a hosted LLM.
Model ladder
Diffio 2.0 for fast, multilingual, high-volume work. Diffio 3.5 for maximum English reconstruction quality.
Evaluation know-how
Specialized data, restoration metrics, listening practice, and a public reproducible benchmark.
Production integration
Model training and inference operated in-house, with API, SDK, webhook, billing, and deployment skill.
Privacy by default
Uploads, outputs, and transcripts do not train shared models without affirmative opt-in. Users retain their content.
NHColorado Springs
10 · Founder-market fit
A decade in signal processing, now pointed at one product.
Nathan Harmon built defense and commercial signal-processing systems, most recently RF geolocation and device identification at Northrop Grumman.
Today he trains models, runs GPU infrastructure, and ships Diffio end-to-end.
Next capability: add commercial distribution and partnership depth around a proven technical core.
11 · Milestones + Gemini Startup Forum fit
Make quality repeatable—in inference, acquisition, and integration.
Operating targets, not financial projections. Google expertise strengthens the system around Diffio’s proprietary restoration core.
- 01
Validate lower-cost, reliable proprietary-model inference and a Google Cloud go/no-go path.
- 02
Turn creator restoration proof into a repeatable acquisition and subscription funnel.
- 03
Secure API design partners and harden workflow, usage, and audit capabilities around them.
- 04
Add the first sustained GTM capability without diluting the restoration focus.
Concrete 90-day outcomeA measured Google Cloud inference decision, one validated complementary Gemini workflow prototype, and qualified partner/customer introductions.
Benchmark proprietary-model inference economics and reliability on Google Cloud.
Test Gemini for metadata, transcript, or QA workflows only where it earns user value.
Meet media-platform partners and design customers with difficult speech at scale.
12 · The ask
Help Diffio turn restoration proof into scaled delivery and distribution.
22.54% greater SCOREQ improvement≈ $9k ARR company-reported, August 2026$0 outside funding
Seeking program partnership, technical mentorship, and introductions—not presenting an unverified fundraising round.