Engineering deep dives, product comparisons, and updates from the DictatorFlow team.
Our new pronunciation game shows your voice overlaid on a native-speed target waveform in real time. Green means you matched the rhythm. Ten languages, no sign-up.
Ten rounds, twenty languages, four choices per clip. Guess the Language trains your ear to tell Portuguese from Spanish and Korean from Japanese.
Shadowing — speaking along with native audio in real time — is the highest-leverage pronunciation technique. Here is the method, and the free tool that gives you instant feedback.
DictatorFlow is moving from dictation into hands-free computer control, with workspace tools for files, search, command execution, and agent delegation.
The DictatorFlow chat endpoint now supports tool-calling loops for local workspace actions, giving live chat a real path from answer to edit.
The local chat tool layer can optionally launch Codex inside a workspace, making voice and live chat a front door for coding agents.
Local computer-use tools are powerful, so DictatorFlow scopes file access to a configured root and rejects absolute paths, parent traversal, and symlink escapes.
Conversation mode, live chat, local TTS, and computer-use tools are converging into a voice-first agent loop for real desktop work.
Speech should do more than fill text boxes. The real productivity jump comes when voice can inspect files, run tools, navigate media, and delegate computer work.
A hybrid transcription router can keep the common path local and instant, then use remote models only when confidence or disagreement says the transcript needs help.
Latency work only counts when accuracy holds. Our optimization loop measures WER, time-to-result, provider failures, chunk fallback, and audio preprocessing together.
The desktop player now indexes local folders for music, movies, and videos, with named media indexes so each library can have its own folder roots.
Our homepage already passed Core Web Vitals. The real problem was that crawlers and link unfurlers saw a blank page. Here is how we measured it, what we fixed, and the tooling we built to keep it fixed.
We looked at multiple launch venues for Dictator. Pump.fun won on reliability, distribution, and time-to-market. We evaluated Bags.fm too, but hit enough bugs in critical flows that we chose not to launch there.
One script tag, one mic button, one waveform modal. The DictatorFlow browser widget is now packaged for open-source release under the lee101 account.
We benchmarked DictatorFlow against Google, Deepgram, Otter, WhisperFlow, AssemblyAI, and Amazon Transcribe on price, accuracy, and latency. The results aren't close.
How DictatorFlow uses online learning, pre-augmentation pipelines, and multi-model ensembling to continuously push word error rates below 1.2%.
From Zig audio processing to Go API servers -- a look at the architecture powering sub-150ms transcription across every platform.
WhisperFlow charges $12/mo for Whisper-based transcription. DictatorFlow delivers lower WER, lower latency, and offline mode for $9/mo.
Otter.ai is built for meeting transcription. DictatorFlow is built for real-time dictation. Different tools for different jobs -- here's why dictation users should choose us.
Deepgram offers great STT APIs for developers. DictatorFlow gives you that plus a complete desktop app, offline mode, and simpler pricing.
Google Cloud STT is powerful but complex. DictatorFlow delivers better accuracy with flat pricing and no GCP console required.
The average person types 40 WPM. The average person speaks 150 WPM. That's not a small gap -- it's a 4x productivity multiplier hiding in plain sight. Here's how real teams are using DictatorFlow to capture it.