80 two countries - speaking the MS Translator

[jutfrank]

I wonder whether you guys can get the correct version - "take it" (When I read "take it or leave it", it correctly shows "take it", of course).

It correctly transcribed take it for me every time I tried.

(Also with provided.)

[GoodTaste]

Really? Interesting.

GoodTaste, do you know if there are settings where you can change the voice of the speaker?

The web version doesn't seem to have the options - it just sticks to US International Version of English.

[GoodTaste]

I wonder whether you native speakers make the Translator recognize your voice correctly when you read aloud "Richard Dawkins' motto: In Science We Trust."

The best result I got is "Richard Dawkins motto: In Science We Trust." That is, I deliberately left out "'s" in "Dawkins's" - according to the grammatical rule, "s's" should be pronounced as "/ziz/". But the translator refuses to recognize me.

[jutfrank]

I get: Richard Dawkins is motto in science we trust.

[GoesStation]

I wonder whether you native speakers make the Translator recognize your voice correctly when you read aloud "Richard Dawkins' motto: In Science We Trust."

The best result I got is "Richard Dawkins motto: In Science We Trust." That is, I deliberately left out "'s" in "Dawkins's" - according to the grammatical rule, "s's" should be pronounced as "/ziz/". But the translator refuses to recognize me.

You're expecting too much from a speech-to-text engine. Be happy with the amazingly good results they often produce.

[Tdol]

But if we compare this to where they were ten years ago, we're complaining about individual words, when they were turning written texts into utter gibberish. They will get there. The rate of improvement has been astonishing.

[GoesStation]

But if we compare this to where they were ten years ago, we're complaining about individual words, when they were turning written texts into utter gibberish. They will get there. The rate of improvement has been astonishing.

We're really talking about speech recognition here, by an engine that happens to be embedded in a translator. Both have improved dramatically. I'm confident in predicting that automated translation will never be as good as human translation. Machines would have to acquire the emotional response that makes a human know that one version is right and another clumsy or ludicrous. But the neural networks that Google uses do a terrific job of approaching human capabilites.

[GoodTaste]

Please read "Reconstruction of the oocyte transcriptional network with transcription factors"* to the Translator. It appears to be easy except the word "oocyte". See how many times you have to take before "oocyte" be correctly spelt out.
The best result I got is:
(Click to enlarge)
I've tried several times and got the same result as above. I pronounced it as ['əʊəsaɪt]. I am eager to see the result of yours.

* It is the title of a paper published in Nature on 16 December, 2020 by Japanese scientists about reversing the aging process.

[GoesStation]

I'd be astonished if a speech-to-text engine got the rare, technical word "oocyte" right unless it was one used in medical transcriptions.

[jutfrank]

It correctly transcribed oocyte for me the first time and every subsequent time.