3 ms·

The issue of intonation, here, isn't a technical one. I did my third years' internship in a 3 people startup, lost in the middle of the french countryside. And

by Fiahil 10y ago

The issue of intonation, here, isn't a technical one. I did my third years' internship in a 3 people startup, lost in the middle of the french countryside. And we were working on a software that could generate complete weather bulletin for any european city (>50k inhabitants) or simply deliver a personalized voicemail (with names, dates, past purchases) like any human would do. In fact people were calling back the number we gave, asking for "jeremy", the bot. It was really good and it was less than 1k lines of python code (embedded in django). So, if Amazon wanted it to be something more than a marketing fluff then could -probably- do something about it.

rlpb 10y ago
Being "really good" may not be enough. If it's not perfect, being "really good" could actually make it worse: https://en.wikipedia.org/wiki/Uncanny_valley https://en.wikipedia.org/wiki/Uncanny_valley
- Fiahil 10y ago
  http://www.creawavestudio.com/interfaces_vocales/ http://www.creawavestudio.com/interfaces_vocales/ Is it good enough for you? :) Note: I worked there in 2013, they are smart people. And yes, these are _computer generated_ messages, with human voices.
  - hectormalot 10y ago
    That's actually pretty neat! So how come other voices don't match this? Is it the fact that intonation in a single domain (weather) is easier to predict than if the voice needs to handle general (broad) topics of conversation?
    - Fiahil 10y ago
      It's not about the topic, but mostly the technique. You have to record and process a few hours of human speech in a studio (when I was there the "studio" was a closet with blankets on the wall). Most of the words are not cut in the same way as others text-to-speech engine would do, to preserve intonation, hence longer recording times and a "grammar" limited to the the span of the recorded vocabulary (see if you can catch how the topic is always about the rain. Maybe it was due to us being in britany, a very rainy french region) However, given the number of complaints here, I don't think a cap on the number of topics she can discuss would be an issue for Alexa.
      - hectormalot 10y ago
        Thanks, insightful to know.
  - eitland 10y ago
    This is nice! German didn't work for me though.

by Fiahil 10y ago

rlpb 10y ago

Being "really good" may not be enough. If it's not perfect, being "really good" could actually make it worse: https://en.wikipedia.org/wiki/Uncanny_valley https://en.wikipedia.org/wiki/Uncanny_valley

Fiahil 10y ago
http://www.creawavestudio.com/interfaces_vocales/ http://www.creawavestudio.com/interfaces_vocales/ Is it good enough for you? :) Note: I worked there in 2013, they are smart people. And yes, these are _computer generated_ messages, with human voices.
- hectormalot 10y ago
  That's actually pretty neat! So how come other voices don't match this? Is it the fact that intonation in a single domain (weather) is easier to predict than if the voice needs to handle general (broad) topics of conversation?
  - Fiahil 10y ago
    It's not about the topic, but mostly the technique. You have to record and process a few hours of human speech in a studio (when I was there the "studio" was a closet with blankets on the wall). Most of the words are not cut in the same way as others text-to-speech engine would do, to preserve intonation, hence longer recording times and a "grammar" limited to the the span of the recorded vocabulary (see if you can catch how the topic is always about the rain. Maybe it was due to us being in britany, a very rainy french region) However, given the number of complaints here, I don't think a cap on the number of topics she can discuss would be an issue for Alexa.
    - hectormalot 10y ago
      Thanks, insightful to know.
- eitland 10y ago
  This is nice! German didn't work for me though.