01
Text from your pipeline
We receive the text as it moves through your system — before or as it becomes an article. No page scraping, no reading the live site.
Article live.
Text in.
Audio ready.
Your systems send the article text as it moves through your pipeline — before or as it becomes a web article. A Mac beside editorial synthesizes listen-audio in Dutch and — via optional auto-translate — in other languages, with a choice of female or male voice. Articles stay in-house: no cloud TTS, no unknown servers. We deliver an audio player, an API, or both — whatever your apps and site need.
Flow
From publish to play
02
On-prem speech
Synthesis on a Mac on your network. Article text does not go to ElevenLabs or other cloud TTS — it stays with you.
03
Player or API
We deliver an embeddable audio player (+ playlist, chapters, optional ad slots on your terms), a publish/status API for your own apps, or both — matched to your stack.
04
Self-learning
The system randomly samples audio, learns from pronunciation and quality, and continuously improves the lexicon and speech.
Article edited? When editorial updates the piece, audio can be regenerated — the same publish step with the new text, so listeners always hear the current version.
Generated audio is yours: freely use and distribute it via website, apps, podcast feeds, smart speakers, and other software — you choose the channel.
Privacy & control
Articles stay in-house
Article text does not leave your office to a third party. No ElevenLabs, no big-tech speech cloud, no servers “somewhere” whose location you do not know. Speech runs on-prem with you — Mac and adapter on your network, under your security and data control.
No third-party TTS
No article text sent to external speech vendors or unknown SaaS.
Known location
Hardware and software sit where you put them — your building, your network, your rules.
You own the audio
MP3s and CDN are yours; you decide distribution to site and apps.
Hardware
Why Mac M4 or newer
Speech runs locally on Apple Silicon next to editorial — not a GPU rack in the cloud. We recommend a Mac mini or Mac Studio with M4 or newer: enough unified memory and Neural Engine for news articles inside your go-live SLA, quiet and efficient enough for 24/7 next to IT.
On-device neural speech
Modern neural voices need unified memory and Apple Silicon acceleration. M4 (and newer) has the headroom for long articles, voice variants, and optional auto-translate on the same box.
Fast enough at publish
At peaks (several lives at once) the queue waits on one active generate. Stronger chips shorten time-to-ready — critical for “article live → quickly listenable”.
Quiet, efficient, beside editorial
A Mac mini/Studio fits a newsroom or IT corner: low heat, low power, no datacenter noise. Always-on queue without exotic AI hardware.
A SKU your IT already knows
Standard Apple gear: MDM, network ACLs, procurement and support you already run. No public port required — only LAN/VPN to the adapter.
Lighter chips can work for a small demo; for production volume and failover we advise M4 or newer. Exact SKU (mini vs Studio, memory) is locked in Discovery from articles/day and peaks.
Delivery
Player, API, or both
Not every publisher wants the same front end. We match your needs: ready-made player (with chapters and optional ads on your rules), API-only for your apps, or the combination.
Audio player
Embed module with play/pause, speed, playlist/queue — your team mounts it in the website or an app webview.
Chapters
Audio can be split into chapters/sections — jump to paragraphs or headlines. Layout follows your wishes (CMS structure or editorial markers).
Ads (optional)
Slots for pre-/mid-roll or markers in the listen experience — only if you want them, on your terms (none, light, or wired to your ad stack).
API
Publish and status API: your CMS and apps request audio, poll until ready, play or cache themselves — including chapter/ad metadata when you use it.
Both
Player for the site, API for native apps, podcasts, or other software — one speech backend.
Voice & language
Translate and choose a voice
Optional auto-translate turns the article into another listen language before speech starts — not only English, but other languages too. Listeners (or editorial) pick a female or male voice — same pipeline, your brand voices or stock voices.
Auto-translate
Source text from your pipeline → translated speech text → audio in the chosen language (English or others).
Female or male voice
Choice per article, per language, or as a site default in the listen module.
Listen
Audio from the demo
Real articles from our demo site — Dutch and other languages. Some demos include an optional sponsor preroll (ads) + chapters, the same kind you can enable yourself.
De belangrijkste plannen van minderheidskabinet-Jetten op een rij
De Brics: een gevaarlijk antiwesters blok of louter een spookvijand?
Zes verdachten opgepakt na vondst van 22.000 kilo vuurwerk
Amerikaanse tiener overleeft dagenlang in kou op omgeslagen boot in Beringzee
For demonstration only. Article text and rights remain with the publisher.
Open demo siteQuality
Sample, learn, improve
After audio generation, the system randomly checks pieces of audio. Findings flow back into pronunciation rules and speech quality — a self-learning system that grows with your newsroom, without listening to every article by hand.
Two layers
Speech and delivery
Speech pipeline
Text from your pipeline → (optional translate) → MP3. All on-prem with you — no article text to ElevenLabs or other cloud TTS. Failover-ready, normalization, voice choice, plus a self-learning loop.
Player and/or API
Embeddable audio player + playlist, with optional chapters and ad slots on your terms — and/or an API that feeds your own player, apps, and other software. Serves only your audio URLs — never scrapes the page.
You supply
- Article text from your pipeline (before/at publish)
- CDN / storage for MP3s
- Integration (template and/or app client)
- Distribution to apps and other channels
We supply
- On-prem Mac speech stack + adapter (at your site)
- Publish/status API & runbooks
- Optional: player + playlist, chapters, ad slots
- Failover: backup Mac or backup drive within the Pilot price
Offer
From Discovery to care
Discovery
More detail
Typically 2–3 weeks from kickoff. Discovery is a fixed fee — not open-ended hours — aimed at a binding Pilot quote. The amount follows after intake.
Included: kickoff (~90–120 min) with product + platform; technical sessions on how text arrives from your system (before or as it becomes an article); volume (articles/day, peaks, text length) → Mac sizing and feasible SLA; player module vs API (or both); failover choice (backup Mac or dedicated backup drive).
Deliverable: short findings (topology, API fit, player sketch, risks) plus a fixed Pilot price. No production install in Discovery; no browser extension that scrapes the page — text comes from your pipeline, not the live page.
Pilot — speech
More detail
Duration 4–8 weeks from Pilot kickoff, depending on network access, Mac delivery, and CMS wiring. Price is fixed after Discovery — no surprise hardware line for the failover unit.
Included: on-prem speech stack on the primary Mac path, adapter/queue, publish and status API, staging and production smoke. Inside that price, failover is guaranteed: either a second Mac or a dedicated backup drive — locked in Discovery.
You usually buy the primary Mac; the backup Mac or drive sits inside the Pilot speech fee. Out of scope unless agreed separately: full CMS rewrites, CDN build-out by us, or 24/7 on-call outside business hours.
Pilot — player
More detail
Extra 2–4 weeks that can partly overlap Pilot speech. Fixed price after Discovery, based on template complexity and branding.
We deliver the embeddable audio player (play/pause, speed, playlist/queue) plus an integration guide; together we mount it on one article template. Language and voice choice (female/male), chapters in the audio, and optional ad slots follow your wishes — you decide if and how ads play.
Prefer your own UI? We supply API docs and examples so your apps or site player only fetch audio URLs, status, and optional chapter/ad metadata — no DOM scrape. You keep ownership of audio and distribution.
Annual care
More detail
Annual retainer that starts after Pilot: updates to the stack and (if purchased) player module, lexicon/pronunciation care, support in business hours. Price on request.
Failover stays in the runbooks (backup Mac or drive from Pilot). No separate surprise for that coverage while care runs — still tied to the hardware choice you locked in.
Outside standard care unless agreed: overnight 24/7 on-call, large features off-roadmap, or formal security certification. Those can be scoped in Discovery or at renewal.
Optional: brand-voice train (one-shot, price on request). Primary Mac usually purchased by you; backup Mac or backup drive is included in the Pilot speech price (choice locked in Discovery).
No price list on the site — after a short conversation you receive a fixed quote tailored to scope, volume, player/API, and failover.
Next step
Discovery kickoff
Ninety minutes with product + platform: text path, player or API, fixed Pilot quote.
Get in touch