Notes from the voice lab
Engineering deep dives, product news and what we learn shipping voices people talk to every day.
Latest,
Testing voice apps like you test code
Voice features break quietly. A practical setup for catching regressions before your users do.
Jordan Wells, Evaluation Lead
What forty languages taught us about pauses
Silence is part of speech, and every language uses it differently. Notes from scaling to forty.
Oliver Grant, Languages Lead
On-device or cloud speech: an honest comparison
Running speech on the device sounds great until you check the battery. When each one makes sense.
Isaac Turner, Platform Engineer
Picking a voice for your brand
Your voice is part of your brand, whether you choose it or not. How to choose it on purpose.
Juliet Ramos, Voice Director
Pronouncing the unpronounceable
Brand names, drug names and surnames break most voices. Pronunciation dictionaries fix them once.
Emma Carter, Linguist
Why your support bot sounds bored
Flat voices are rarely about the voice. They are about missing context. Here is how to give it back.
David Lee, Conversation Designer
Streaming speech over WebSockets, step by step
A plain walkthrough of the streaming API, from opening a socket to playing the first chunk.
Samuel Ortiz, Developer Advocate
Writing for the ear, not the eye
Copy that reads beautifully can sound awful out loud. A few habits that make scripts land.
Maya Lombardi, Voice Content Writer
Voice cloning without the creepy part
Cloning a voice takes thirty seconds of audio. Doing it responsibly takes a little longer.
Leila Fernandez, Trust and Safety Lead
How we got first audio under 200 milliseconds
The boring engineering behind the moment a voice starts talking before you notice you asked.
Marco Bianchi, Head of Speech Research
Subscribe, unless you hate fun
One email a month with new posts, model releases and the latency numbers we are proud of. No spam.