4 ms·

Wow text-to-speech doesn't seem to have gotten very far in the last 20 years? When we were kids we used to use a windows (3.1?) program for prank calls and it d

by ian0 8y ago

Wow text-to-speech doesn't seem to have gotten very far in the last 20 years? When we were kids we used to use a windows (3.1?) program for prank calls and it didn't sound much worse.. I would have thought by now basic things like word spacing and tones would have improved to the point where you would only realise its artificial occasionally.

wenc 8y ago
Not sure if they are using state-of-the-art TTS. This one [1] sounds a bit more natural. [1] https://neospeech.com/ https://neospeech.com/
- Mathnerd314 8y ago
  Still sounds stilted. Try Google's: https://cloud.google.com/text-to-speech/ https://cloud.google.com/text-to-speech/ (https://www.theverge.com/2018/3/27/17167200/google-ai-speech-tts-cloud-deepmind-wavenet https://www.theverge.com/2018/3/27/17167200/google-ai-speech...)
  - AWildC182 8y ago
    Kind of uncanny valley. The pronunciation of the words is good but there's no emphasis through the sentence so it sounds flat and kind of unsettling.
    - candiodari 8y ago
      They should make it a university teacher. Clearly it's got the core skill down.
  - gambler 8y ago
    Try to enter a paragraph of text you didn't see before and close your eyes while it's "speaking". Do not read along. It does not have a lot of obvious artifacts that plagued older TTSes, but it actually is remarkably hard to understand.
MaBu 8y ago
It definitely doesn't use state of the art TTS. It sounds similar to MBrola english voice. There is google's [Wavenet](https://deepmind.com/blog/wavenet-generative-model-raw-audio/ https://deepmind.com/blog/wavenet-generative-model-raw-audio...) and even chinese Baidus [Clarinet](https://clarinet-demo.github.io/ https://clarinet-demo.github.io/) which are much better.
- ian0 8y ago
  Cheers - yep google's especially sounds much more realistic. Obviously still awhile to go before a user would be tricked into thinking it's a real voice.
reaperducer 8y ago
Wow text-to-speech doesn't seem to have gotten very far in the last 20 years FWIW, I find that the macOS' Tom voice is remarkably good. I sometimes wonder if it's what NPR uses for some of its on-air messages, or if Apple's voice was just based on the same guy.

by ian0 8y ago

wenc 8y ago

Not sure if they are using state-of-the-art TTS. This one [1] sounds a bit more natural. [1] https://neospeech.com/ https://neospeech.com/

Mathnerd314 8y ago
Still sounds stilted. Try Google's: https://cloud.google.com/text-to-speech/ https://cloud.google.com/text-to-speech/ (https://www.theverge.com/2018/3/27/17167200/google-ai-speech-tts-cloud-deepmind-wavenet https://www.theverge.com/2018/3/27/17167200/google-ai-speech...)
- AWildC182 8y ago
  Kind of uncanny valley. The pronunciation of the words is good but there's no emphasis through the sentence so it sounds flat and kind of unsettling.
  - candiodari 8y ago
    They should make it a university teacher. Clearly it's got the core skill down.
- gambler 8y ago
  Try to enter a paragraph of text you didn't see before and close your eyes while it's "speaking". Do not read along. It does not have a lot of obvious artifacts that plagued older TTSes, but it actually is remarkably hard to understand.

MaBu 8y ago

It definitely doesn't use state of the art TTS. It sounds similar to MBrola english voice. There is google's [Wavenet](https://deepmind.com/blog/wavenet-generative-model-raw-audio/ https://deepmind.com/blog/wavenet-generative-model-raw-audio...) and even chinese Baidus [Clarinet](https://clarinet-demo.github.io/ https://clarinet-demo.github.io/) which are much better.

ian0 8y ago
Cheers - yep google's especially sounds much more realistic. Obviously still awhile to go before a user would be tricked into thinking it's a real voice.

reaperducer 8y ago

Wow text-to-speech doesn't seem to have gotten very far in the last 20 years FWIW, I find that the macOS' Tom voice is remarkably good. I sometimes wonder if it's what NPR uses for some of its on-air messages, or if Apple's voice was just based on the same guy.