Hacker News Reader

Comment by whalesalad | Hacker News Reader

The front-and-center 'voiced by AWS Polly' dogfooding is great but it sure does sound like crap. Ironically it struggles to pronounce the word Milan.

View on HN

I am happy to share feedback with the Polly team! Can you be more specific about what you believe sounds bad?

Other than the "milan" pronunciation issue that others have outlined: the same word is pronounced in exactly the same way every time, which sounds wrong especially in long words such as "availability". The speed is too uniform; Polly handles commas (and full stops) brilliantly but there are other occasions in which a huan speaker pauses ever so slightly, for example to take a breath. Humans don't pronounce every single word fully, which contribute to make speech sounds fluid, this seems to be lacking completely from Polly.

When humans read aloud, they usually sound pretty bad. We tend to read sentences word-by-word instead of breaking it down into parts. I have no idea how Polly actually works, but it sounds to me like it is reading word-by-word like many humans do.

I'm (probably overly) conscious about how I read things aloud. If I can't read a sentence quietly before I read it out loud, I try to diagram it as I go. For instance, if I identify what I'm reading as a prepositional phrase, I give it less vocal emphasis. Subjects and predicates usually get the most emphasis.

Of course, what I described is much easier said than done. I wouldn't be surprised if you were already doing it.

Good luck with Polly moving forward and thanks for being so willing to take user feedback!

For me constantly gasp before sentences sound bit weird. It would be better if this occurs only after long sentences, would sound more natural.

Nevertheless sounds really good!

I thought the same. It would probably sound more realistic without the gasping. Most human speakers wouldn't gasp that audibly and at different parts of the text.

It would be really great if the general intonation can be improved to be nearer to Wavenet level. Also a lot of continental Germanic words e.g. Swedish, Dutch, Deutsch etc. and Français-derived words embedded in an English text won't be pronounced properly regardless of context.

Source: I run an ebook to audiobook converter https://auditus.cc

for what it's worth I thought it sounds fine. Maybe Milan sounds wrong to an Italian, but it sounds like how most Americans say Milan to me

Yeah, I think you're right. I suspect the Italian pronunciation is like 'me-lan' rather than 'mil-an'. As for the general sound, I thought it was pretty good. I was impressed that Polly even adds in inhaling of breath. Subtle, yet effective.

Italian pronunciation for Milano is here: https://www.comesipronuncia.it/pronuncia/milano-10397 . That said, Italians generally mean the UK’s received pronunciation when studying and listening to English language, so heavy American accents are way too large and quite difficult to us.

After going back and forth between the article and the comments, I just realized that you're the author / Chief Evangelist at AWS.

So on a sort of meta-note, I think it's pretty cool that AWS is still interacting with developers at this level.

It is not too bad, but is not 'Maelan' but 'Meelan'.

It pronounces Milan perfectly to my British ear. It would be pronounced much the same here minus the accent. Certainly not Meelan.

I was talking about the italian pronuntiation.

Well the English word "Milan" is already mangled compared to the Italian "Milano" - taking the italian pronunciation for the English word would be ... odd. I'm pretty confident that in the anglosphere "Mee-lan" would be the minority pronunciation compared to "Mih-lan".

Milan Kundera pronunciation comes neat though https://www.pronounceitright.com/pronunciation/milan-kundera... , that’s how we are used over here.

Maybe it's because the name is Czech and we never anglicised his name? I honestly don't know with that one though. English is wildly unpredictable, if we drilled down deep enough we'd find plenty of counter examples that contradicted my reason for how we pronounce pan Kundera's first name :-)

However for the city, think of it this way (sorry I'm gonna use some awful phonetic spelling here!) you and all the other Italians use the 100% correct pronunciation of Meelano. Somewhere back in History it became normal to call it "Milan" in English. Calling it Meelan is not only still wrong in Italian, it's also wrong (or nearly unused) in English.

It's the same way you have your own word for our city with its own pronunciation - Edimboorgo. If I say "well, that 'u' in the original 'Edinburgh' is pronounced uh so you should call it Edimbuhrgo" then you're now pronouncing it wrong in both Italian and English.

For what it's worth I'd prefer it if English kept the original names - Milano, Torino, Roma, Napoli, Firenze, etc :D

Edit I realised after typing this that I've heard wealthier Americans pronouncing some of these names even stranger still. Milan becoming Milawwn and Prague becoming Prawwwg which is like nails on a chalkboard to me, so maybe I can identify with you a little bit :-)

Ehehe, I see you well: let’s agree Mee-lah-n would be the best and neatest compared to received pronunciation and move on happily.

Why not :D

In defense of the software, Italian is quite difficult to pronounce. I wouldn't be too hard on it for now.

Disclaimers: I am Italian, and I used to work at AWS.

It's one of the simplest language to pronounce, along with Spanish

I'm surprised, how is Italian difficult to pronounce? I'm Italian too. I understand if you mean difficult for somebody who has 0 knowledge (this is true for any language). But besides very few exceptions Italian pretty wysiwyPronounce, no? I feel like I'm blinded by being too close to it :)

As an Italian expat: accent positions are very, very different; and English people associate different "default" sounds to vowels in isolation (e.g. "a").

Even educated persons will struggle if their only source is paper, because a lot of the Latin words that Italian is rooted on, actually have very different pronounciations in English.

Also: GL, GH, GN, CH, CE/CI vs CA/CO/CU, double-consonants reinforcing the sound, etc etc...

Most Italian words pretty much match their IPA representation, minus the pretty regular exceptions like Q, C, G, Z and H and the rare irregularity.

It may be difficult to an anglophone, but I'd argue that the phonetic irregularities of English are much more of a problem to a learner. "though/tough/through" never happens in Italian.

The biggest problem with reading Italian is knowing where the accent goes if it's not mandatory to write, in which case you need to know from context, as in the words at the end of http://www.italianlanguageguide.com/pronunciation/stress-acc...

agreed, the accent location (stressed syllabe) comes down to knowing the word, but only a few cases of wrong accent give you a different word... in english you can change the word by mispronouncing the lenght of the first vowel in words like sheet, beach... both long vowel, different way to write it... although I'm sure I'm being imprecise on "long vowel" (don't know anything about phonetics), but I can't think of an example of non-accent based, italian equivalent

on the accent, fair enough, that is not straightforward.

On the default sounds: I think it's harder in english where the correct sound is not strictly connected to the letters you write the word with. I don't think it's a coincidence that spelling bees are a thing for the english language, but I never heard of one in Italy...

Sure, if you have no source to determine the pronunciation of a word, it's hard (as for any language), but once you learn the building blocks you don't strictly need to know the word to know the pronunciation (apart from the accent)

Imho it speaks way too fast.

Reply on news.ycombinator.com