🧵 View Thread
🧵 Thread (19 tweets)
"AI" is just some guy behind a curtain listening to you take a dump, news at 11.https://t.co/ELWps68umu
"This is not the first time we've been forced to do a collective double take upon learning how supposedly smart tech actually works."https://t.co/LMLfUY1AcL
I have my echo dot set to "echo" as the wake word. A guy coughing out my window has set it off before.My guess is what's happening: cloud voice rec systems generate guesses at what was said, use human-in-the-loop to quickly select from choices when the confidence is low.
"still has many challenges to address, such as achieving human levels of recognition in noisy environments with distant microphones, in recognizing accented speech, or speaking styles and languages for which only limited training data is available."https://t.co/DiXH6JUFp1
So go to https://t.co/tuh3B1XrrfSelect "Manage How Your Data Improves Alexa" on the navigation menu (left?)Under "Use Messages to Improve Transcriptions" uncheck next to all names listed.Now your voice transcripts won't be listened to to customize them for your voice profile.
There is also an option for "Help Develop New Features" that may be similar.Worth reviewing https://t.co/uo1AXJsXbQ which covers their privacy policy for alexa.
If you are surprised by how all of this works. You haven't been paying attention.https://t.co/9sGdsydvYI
You can review google's help page that details how they store and use your voice transcriptions: https://t.co/6IdStZ0d27In particular, you may want to review "Turn Voice & Audio Activity on or off" and "Delete recordings from your Voice & Audio Activity".
Theoretically you can minimize error rate of your voice transcription by ensuring your {amazon alexa, google home, apple siri, microsoft cortana, samsung s voice, comcast X1 remote} is close to you, & away from sources of noise like fans or television. Welcome to the cloud baby!
Market Research. Both cost me $0.00.Only thing I haven't tried is Siri.https://t.co/dEpFbIAkOi https://t.co/X6yi4mMvaz
"gotta make it yourself."https://t.co/MHi2Dekbez
"We have recently launched an open-source repository on GitHub, where we are rapidly iterating on the global open standard, based on market-proven technologies. "https://t.co/vDsMrAL0ASTL;DR: ignoring all the small players and only targeting industry insiders who donate.
I can confirm that if you screech just right, the echo understands me.https://t.co/CJdKf1W1bT
Teaching myself clojurescript so I can interface with JavaScript based voice APIs (like the Web Speech API) with a bit of a lisp.https://t.co/tsnXhLd9RwGoing to teach my computer to recognize aural conlangs by leveraging word-vectors as metaphor resonance induction.
Hello computer.I want to {Bed, Pillow, Night, Snore, Dream} =>"Ok, I'll activate the sleep routine!"Lethologica makes interacting with voice assistants nearly impossible sometimes. Stan'ing myself.https://t.co/8zgKLbH5uw
"The 55-year-old was told he’d never speak again following life-saving surgery but defied his prognosis by learning to make sounds in a totally different way, a skill which is rarely mastered."https://t.co/eV0l5KOuRE*raspberry tart noises*https://t.co/dgIDR9mDcp
The lovely people at @FUTO_Tech have given me the ability to use my phone without needing to send my voice transcripts directly to google.https://t.co/LfZFV2iA6j
Scribe of the gods.https://t.co/movIe3wEIn
I found some old notes today from 2013 where I basically predicted how to implement the hybrid human-ai centaur system behind Google Assistant, Alexa, and Cortana.I was going to call itThoth Transcription Services, LLC.https://t.co/sKEJhU6YXS https://t.co/h1gXEL2OVR
