4 ms·

This is actually going to remain a problem until computerized speech-to-text is solved. And I'm pretty sure speech-to-text will not really be solved until gener

This is actually going to remain a problem until computerized speech-to-text is solved. And I'm pretty sure speech-to-text will not really be solved until generalized artificial intelligence is solved, so this will be quite some time.

praptak 15y ago
Even with perfect room-of-people-interrupting-each other text-to-speech, there would remain one important use case - writing down own thoughts. Maybe also descriptions of live events where a lot happens at once?
scotty79 15y ago
My thoughts exactly. To understand what was said you often have to understand the context and the intention. Otherwise you have no chance at decoding garbled sounds people make even with you carefully optimized meatware. Perfect text to speech also needs true AI that understand the text it is reading it's context and intention. Reader must understand the message to read properly.
- ploversteno 15y ago
  Thank you! It's so refreshing to talk with coders about this stuff; non-computery types so often assume it's a solved problem, because they don't realize how lossy a format spoken language actually is.
Vivtek 15y ago
According to what I'm reading, speech is only about 180 WPM and steno can be 240-300 WPM. Speech-to-text will actually be less efficient than steno. And for me, typing 1.25 million words of English per year (in technical translation, mostly from German) - if I tried using speech-to-text my larynx would be burnt out in a month. No way. I've heard some people get mileage out of it, but not me. I'm learning me some steno, starting today.
- fragsworth 15y ago
  I can't tell if you're joking or not. The point of speech-to-text is not necessarily efficiency, it's also for subtitling and/or recording normal speech (i.e. phone conversations, or TV screens in public areas)
  - Vivtek 15y ago
    Ah. Yeah, for that it's obviously going to be the bee's knees except for the whole AI-completeness issue. But for me it's useless; I look at German text and type English text, so the issue really is throughput. (Mostly.) So no, not joking at all. Just talking about a different application.

praptak 15y ago

Even with perfect room-of-people-interrupting-each other text-to-speech, there would remain one important use case - writing down own thoughts. Maybe also descriptions of live events where a lot happens at once?

scotty79 15y ago

My thoughts exactly. To understand what was said you often have to understand the context and the intention. Otherwise you have no chance at decoding garbled sounds people make even with you carefully optimized meatware. Perfect text to speech also needs true AI that understand the text it is reading it's context and intention. Reader must understand the message to read properly.

ploversteno 15y ago
Thank you! It's so refreshing to talk with coders about this stuff; non-computery types so often assume it's a solved problem, because they don't realize how lossy a format spoken language actually is.

Vivtek 15y ago

According to what I'm reading, speech is only about 180 WPM and steno can be 240-300 WPM. Speech-to-text will actually be less efficient than steno. And for me, typing 1.25 million words of English per year (in technical translation, mostly from German) - if I tried using speech-to-text my larynx would be burnt out in a month. No way. I've heard some people get mileage out of it, but not me. I'm learning me some steno, starting today.

fragsworth 15y ago
I can't tell if you're joking or not. The point of speech-to-text is not necessarily efficiency, it's also for subtitling and/or recording normal speech (i.e. phone conversations, or TV screens in public areas)
- Vivtek 15y ago
  Ah. Yeah, for that it's obviously going to be the bee's knees except for the whole AI-completeness issue. But for me it's useless; I look at German text and type English text, so the issue really is throughput. (Mostly.) So no, not joking at all. Just talking about a different application.