Learn with BotTalk · Notes from the orchestration layer · 2026
One control layer.
One operating playbook.
Notes from inside.
What we ship, why we ship it, and what we’re learning from running text-to-speech for publishers at scale across thirty European newsrooms.
The articles
Pick the piece that’s closest to where you’re standing.
-
German voices, made in Germany.
SONA is now in BotTalk.About three quarters of the audio we produce is German. SONA was built in Potsdam by a team with roots at Fraunhofer IIS, the institute where mp3 comes from, and it is now one of the voices in BotTalk. Who built it, what you can set, and what we cannot tell you yet.
Read the article -
BotTalk as a ReadSpeaker alternative.
A side-by-side comparison
for newsrooms.ReadSpeaker brings a proprietary TTS stack with a long accessibility record. BotTalk routes article audio across five external voice providers and adds the publisher business layer: monetization, article updates, German hosting. The architectural question is whether the newsroom wants one TTS vendor or a control layer across multiple providers.
Read the comparison -
Privacy-first voice AI for publishers.
The 2026 procurement case
under GDPR, EU AI Act, and SOC 2.GDPR’s €20M fine ceiling, EU AI Act Article 50’s synthetic-audio disclosure duty from 2 August 2026, and a Data Processing Agreement the buyer’s legal team will read line by line. Privacy stopped being the footnote — it’s the argument that closes the voice-AI deal.
Read the article -
AI voice provider single-vendor risk.
Why every newsroom needs
a control layer.The single-vendor bet keeps failing publishers — outages, price hikes, model deprecations. The fix isn’t a better vendor. It’s a control layer that routes across providers, fails over on the fly, and keeps editorial quality above every model change underneath.
Read the article -
Best text-to-speech for news sites.
The honest 2026 comparison.Every “best TTS” listicle ranks its own product first. Here’s the field described fairly — ElevenLabs, Google, Polly, ReadSpeaker, BeyondWords, Murf, Trinity — sorted into three categories, with the one question every listicle skips: one AI voice engine, or five?
Read the article -
Your competitor isn't a paper.
It's the feed.
Audio wins what's left.Editors benchmark their newsroom against the paper across town. The reader doesn’t. The reader has nearly seven hours online, forty-seven seconds of patience, and an infinite feed. Audio wins the background hours the feed can’t reach — the commute, the kitchen, the gym, the school run.
Read the article -
Five ways to fuel long sessions
with audio articles.
Forty-seven seconds, four-minute listens.Average attention on any screen is 47 seconds. Audio articles hit 75% completion across thirty European publishers. Five operational levers that compound that gap into longer sessions — front-loaded player, cloned voices, voice-to-story orchestration, podcast feed, pre-synthesis quality engine.
Read the article -
Protect content
from LLM training.
The hidden risk in every AI voice pipeline.Every AI voice pipeline runs your articles through two or three LLMs before a single word is narrated. Each stage is its own leak surface. Why content protection matters in 2026 — and the patent-pending obfuscation layer that closes the leak.
Read the article -
Text to speech for publishers.
Why newsrooms are leaving
single-vendor TTS for orchestration.The way text to speech for publishers gets bought has changed twice in eighteen months. First every newsroom ran a pilot on one provider. Then they stopped buying providers and started buying orchestration. Why the shift — and how to evaluate the control layer for your newsroom.
Read the article
More notes shipping every other week. Want one in your inbox? cal.com/bottalk
Older posts, in the blog's original design: Blog archive
Run the play in your newsroom
Talk to Andrey.
No deck. No pitch. Your CMS, your audience, your editorial voice. Thirty minutes. One call.