3 ms·

The stress and intonation examples are what puts this on the far side of the uncanny valley. I've not heard computer speech like this, where the voice seems und

The stress and intonation examples are what puts this on the far side of the uncanny valley. I've not heard computer speech like this, where the voice seems understand the idea underneath the words.

max_ 10y ago
How is this different from the DeepMind's [0]WaveNet? [0]https://deepmind.com/blog/wavenet-generative-model-raw-audio/ https://deepmind.com/blog/wavenet-generative-model-raw-audio...
- stevenhuang 10y ago
  Quoting the paper, in the section related works: "WaveNet (van den Oord et al., 2016) is a powerful generative model of audio. It works well for TTS, but is slow due to its sample-level autoregressive nature. It also requires conditioning on linguistic features from an existing TTS frontend, and thus is not end-to-end: it only replaces the vocoder and acoustic model. Another recently-developed neural model is DeepVoice (Arik et al., 2017), which replaces every component in a typical TTS pipeline by a corresponding neural network. However, each component is independently trained, and it’s nontrivial to change the system to train in an end-to-end fashion."
red75prime 10y ago
It is without doubt very impressive, but on longer texts the lack of understanding will be apparent. Wrong semantic stresses, incorrect pronunciation of homographs, maybe inconsistent pronunciation of names, and so on. Perfect speech synthesis from unannotated text will require much more complex system. It will need database of common knowledge in some form, ability to relate text to that database, ability to retain and use this semantic information to guide synthesis. That is, it should be a step closer to artificial general intelligence.
- jerf 10y ago
  I'd say even in these samples there's still some errors. The intonation on "brown" in "quick brown fox" is still wrong. "AllHipHop" is not pronounced as a human would, though I'll grant you that's a made-up word so I wouldn't even expect a non-native human speaker to necessarily get that one right. Still, the quality is such that I'm only observing this in the spirit of scientific analysis. It's some good quality.
  - iamed2 10y ago
    I wonder whether it would have been more natural if it was written as "quick, brown fox" as would be customary English. Usually when adjectives don't have commas between them it's because the last adjective is coupled to the noun.

The stress and intonation examples are what puts this on the far side of the uncanny valley. I've not heard computer speech like this, where the voice seems understand the idea underneath the words.

max_ 10y ago

How is this different from the DeepMind's [0]WaveNet? [0]https://deepmind.com/blog/wavenet-generative-model-raw-audio/ https://deepmind.com/blog/wavenet-generative-model-raw-audio...

stevenhuang 10y ago
Quoting the paper, in the section related works: "WaveNet (van den Oord et al., 2016) is a powerful generative model of audio. It works well for TTS, but is slow due to its sample-level autoregressive nature. It also requires conditioning on linguistic features from an existing TTS frontend, and thus is not end-to-end: it only replaces the vocoder and acoustic model. Another recently-developed neural model is DeepVoice (Arik et al., 2017), which replaces every component in a typical TTS pipeline by a corresponding neural network. However, each component is independently trained, and it’s nontrivial to change the system to train in an end-to-end fashion."

red75prime 10y ago

It is without doubt very impressive, but on longer texts the lack of understanding will be apparent. Wrong semantic stresses, incorrect pronunciation of homographs, maybe inconsistent pronunciation of names, and so on. Perfect speech synthesis from unannotated text will require much more complex system. It will need database of common knowledge in some form, ability to relate text to that database, ability to retain and use this semantic information to guide synthesis. That is, it should be a step closer to artificial general intelligence.

jerf 10y ago
I'd say even in these samples there's still some errors. The intonation on "brown" in "quick brown fox" is still wrong. "AllHipHop" is not pronounced as a human would, though I'll grant you that's a made-up word so I wouldn't even expect a non-native human speaker to necessarily get that one right. Still, the quality is such that I'm only observing this in the spirit of scientific analysis. It's some good quality.
- iamed2 10y ago
  I wonder whether it would have been more natural if it was written as "quick, brown fox" as would be customary English. Usually when adjectives don't have commas between them it's because the last adjective is coupled to the noun.