Blog
Guides, comparisons, and deep dives on privacy-first transcription, voice-to-text productivity, and the future of local AI.
How VoxBar™ built 15 engines on consumer hardware, tested 8 head-to-head under real-world conditions, and discovered what the benchmarks don’t tell you. From Whisper to Voxtral, Canary to Kyutai — we ranked them all so you don’t have to.
Read the full report →A core team of roughly 100 people shipped over 100 frontier AI models in two years — outpacing labs staffed by thousands. Now their leader has resigned, but their open-source legacy lives on.
Read the full story →We tested INT8, NF4, and F16 configurations for streaming speech-to-text on consumer GPUs. Some paths looked promising until we pushed them. Here's the full breakdown of our quantization experiments and why F16 won.
Read the full report →Meet Pete Warden and Manjunath Kudlur — TensorFlow founding members who built Moonshine, the speech model that runs on anything.
Read more →Meet the engineers whose open-source speech models power three of VoxBar's six engines.
Read more →Meet the team who built the language model backbone that gives VoxBar AI its contextual intelligence.
Read more →Meet the brilliant Paris team behind the Vox Bar Pro engine. From three researchers to a $14 billion company.
Read more →The French open-science lab that cracked real-time streaming speech-to-text.
Read more →How OpenAI's release and SYSTRAN's optimization created the engine that started the offline transcription revolution.
Read more →Your voice contains biometric data. Here's what happens when you upload it to cloud services — and why local AI is the safer alternative.
Read more →Dictate scripts, transcribe voiceovers, take notes during editing — privately, offline, and without monthly subscriptions.
Read more →A new startup is forging permanent AI models into physical chips. Here's how it could put VoxBar in your microphone.
Read more →It's not a brain, a spy, or connected to the internet. It's compressed knowledge in a file. Here's what local AI actually is.
Read more →Don't have a $2,000 GPU? Hardware inference accelerators and open-source models are bringing frontier voice AI to emerging economies in a $50 USB stick.
Read more →Cursor, Copilot, Claude Code — they all make you type. Overlay Mode lets you talk to any AI code editor instead.
Read more →AI understands word relationships, not individual letters. For many people, speaking is genuinely more accurate than typing.
Read more →Microsoft put AI in Office but forgot the microphone. Overlay Mode lets you speak to Copilot, Claude in Excel, and more.
Read more →Download and run AI locally in 5 minutes. No cloud, no API key, no subscription. A beginner's guide to Ollama.
Read more →Ranked: the best AI models that run on consumer GPUs. From Llama to DeepSeek to Mistral — what to use and how much VRAM you need.
Read more →DeepSeek's Mixture of Experts breakthrough made AI models smaller, faster, and good enough to run at home. Here's what happened.
Read more →The hottest AI models this month — from GLM-5's 744B reasoning engine to tiny on-device runners. What's worth downloading.
Read more →AI models are shrinking to fit your laptop — by design. Here's why on-device AI is the future of privacy and performance.
Read more →