Apple fixes Siri’s robotic-sounding voice with deep learning – hear the difference

“Ahead of the launch of iOS 11 this fall, Apple has published a research paper detailing its methods for improving Siri to make the voice assistant sound more natural, with the help of machine learning,” Abhimanyu Ghoshal reports for TNW.

“Beyond capturing several hours of high-quality audio that can be sliced and diced to create voice responses, developers face the challenge of getting the prosody – the patterns of stress and intonation in spoken language – just right,” Ghoshal reports. “That’s where machine learning comes in. With enough training data, it can help a text-to-speech system understand how to select segments of audio that pair well together to create natural-sounding responses.”

“The results speak for themselves (ba dum tiss): Siri’s navigation instructions, responses to trivia questions and ‘request completed’ notifications sound a lot less robotic than they did two years ago,” Ghoshal reports. “You can hear them for yourself at the end of this paper from Apple.”

Read more in the full article here.

MacDailyNews Take: Listen to the iOS 9 and iOS 10 samples vs. the new iOS 11 samples. The new Siri voice is a vast improvement!

SEE ALSO:
Deep learning for Siri’s Voice: On-device deep mixture density networks for hybrid unit selection synthesis – August 23, 2017

21 Comments

    1. “She’d be worse then her husband as the president…”

      That statement is missing a comma so let me fix that for you: “She’d be worse, then her husband as the president…” Perhaps you meant, “She’d be worse than her husband as the president”?

  1. Wow, it’s frighthen how natural it sounds, sky net is closer every day, and the whole explanation its very intresting. Well google and samsung, there you have how they did it, you can start trying to make your cheap imitations for the intellectually poor people.

  2. I like the Siri voice in iOS 10 better then iOS 11. In 11 Siri is too his girl Friday. She is like an eager young intern. In iOS 10 the voice is more stable and emotionally secure. Less eager beaver.

    1. Yes, I wish they’d hired the actress that provided the material for the previous versions, but with the updated intelligence code. It would also be easier to compare the three samples.

    1. Totally agree. The voice improvement is nice. But I’d rather have those scientists work on the way Siri understands questions. Let’s be honest — the Google voice assistant is much better at getting answers than Siri.

      1. I wouldn’t know about that…I do not use Google because I do not trust Google. Google’s products might be wonderful, but I value my privacy far more than I value the utility of Google’s products and services. Besides, I believe that Apple will pull ahead again sooner or later.

      2. I also wish Siri would actually read responses. Often it just pulls up a web page to read when the answer should be something it could read aloud. I also would like the option to have it read text messages aloud. You can do that in CarPlay, but the normal version is just dumb. They also need to teach it things like what’s on TV.

    2. Yes, Siri needs some “deep learning” in her head. There was never any issue with her voice. I don’t care how good she sounds/looks…if she’d dumb, spending the time together is tough.

      1. Bit hard with the amount of data available to Siri compared with her competitors. Apple’s future dedicated AI chip may help but Google is already countering with an onboard lite version of their TensorFlow in ‘Oreo’ as well as opensourcing it to accelerate broader development. A single AI ‘OS’ for all sizes of platforms if you will.

  3. The 11 voice is very bubbly and personal. Kind of like the bubbly gal who calls me several times a month trying to sell me time-share points. She drops her headphones but keeps on talking. I can’t shut her up even when I told her my son was arrested for dealing drugs to minors and the dog died.

  4. Solves one issue I’ve had for a while. Will be more natural using Siri. Now can I please use her to turn on and off my Hue lights without the request having to go to Apple first???

  5. iOS 9 vs iOS 10 female Siri voice: WONDERFUL improvements.

    A) Even more lifelike voice.
    B) No longer sounds like Madonna (IOW it no longer gags me).

    In historical futurism, ‘robot voices’ were monotonic. When the original Mac (1984) could talk better than monotonic, it was considered remarkable. Microsoft attempted something similar and came up with shite voices that sound like Dr. Stephen Hawking. (Sorry Doc). I was personally most pleased when the original Victoria voice was released. *love* But then Apple mangled it in an attempt to make her sound more animated. *fail*

    The best third party voices, IMHO, are from Acapela Group. I particularly enjoy their Peter, Rosie and Queen Elizabeth voices. 👑

Reader Feedback

This site uses Akismet to reduce spam. Learn how your comment data is processed.