Ask HN: Who makes the most advanced text-to-speech technology?
6 comments
Have you tried AWS Polly? https://aws.amazon.com/polly/
Seconded, AWS Polly should be good, there's also Bing Speech
https://azure.microsoft.com/en-us/services/cognitive-service...
Google WaveNet promises even better audio, but it's not available yet. In my experience, Amazon is "good enough".
https://deepmind.com/blog/wavenet-generative-model-raw-audio...
https://azure.microsoft.com/en-us/services/cognitive-service...
Google WaveNet promises even better audio, but it's not available yet. In my experience, Amazon is "good enough".
https://deepmind.com/blog/wavenet-generative-model-raw-audio...
I just gave it a go. It does a good job, but still the pacing and intonation are unnatural and difficult to follow. If this is the current state-of-the-art, or close to it, I would rather just read. It is possible I could train myself to listen to it.
Hats off to the developers, though. I know human-like text-to-speech is difficult.
Hats off to the developers, though. I know human-like text-to-speech is difficult.
although, not a commercially available product, the SOTA for text-to-speach just came out this week and it's called TacoTron 2 https://google.github.io/tacotron/publications/tacotron2/ind...
This might not be very helpful if you are looking for specific texts to be read, but it might be worth trying to find human spoken texts (audio books and so on). Even some medium articles have spoken copies, if I remember correctly.
I do use audio books, and audio versions of articles when available.
I have tried a few text-to-speech offerings, such as the ones included in Firefox and macOS, but they sound robotic to the point that they are difficult to listen to. The pacing is unnatural. I am hoping there is something available that is better.