Espeak has a lot of languages and dialects, possibly more than any other synthesizer. from Common languages like english and spanish, to languages like nahuatl, the aztec language. So for those who speak one of the languages supported by espeak, i’m curious. How accurate are sound and pronunciation in the languages. The only thing I can really make a judgment on is that the American English does not sound quite right.
It really depends on the language and how much work has been done.
Yep, the american english doesn't sound quite accurate. However, it's funny. The british english sounds great and some people are using other languages.
Some languages have had little or no work where the guy talked to a native speaker. A good example of this is Icelandic, my native language. It has had some work, but it still sounds quite funny. However, I use it. It also depends on how good you are at understanding speech synthesizers. I, for example, am quite good at it and so I can use Icelandic eSpeak with the speech rate cranked
up to 90% (99 in the newest nvda versions). SOme people find eSpeak very hard to understand. I personally love it and think it is one of the best synthesizers in the world for good speech rate and responsiveness. But as I say, it's a matter of opinion.
Hope that helps,
#3 (edited by an idiot 2018-11-29 23:08:38)
In some languages, you can tell they synthesize some sounds to replace real sounds. I hear this in Nahuatl, the tl is supposed to make a different sound. The same sound that espeak uses, a fake t, as I have taken to calling it is in latin american spanish. It doesn't really sound like it belongs. Wile I cannot speak the language, I have heard it all my life. The sound, a depressed T of sorts, is very prevalent. When it comes to older speech synthesizers, many things including different voices are synthesized. It’s a pretty safe guess that this extends to some other languages.
the arabic espeek is really bad
Estonian language is also one of those in need for more work. While with some voice variants it suffers from that wrong letter syndrome, with other variants it's actually even usable, but still not perfect.
Polish is good, despite some weirdly pronounced words and acronyms, it's the best polish synthesiser IMO.
Raduvay se domaki ne
Kolko liste po gorach
Tolko zdrave na taz kyshcha
I remember Arabic. Even after they fixed a glitch with it, it’s still doesn’t sound right.
Persian is quite good, though sometimes it pronounces some words wrong. But in comparison to other Persian TTSs, it's still one of a kind. Though the sound is a bit robotic and the quality isn't satisfying, it's still usable.
Other Persian TTss are usually too slow, have poor quality, and or even are generally not worth the high price.
As far as with the German TTS, I can say it mispronounces some words (Though I compared it with the Eloquence voice). I'm not very good at this language, but I think it still needs some work.
Czech is quite good. Of course, Espeak is always going to be robotic, but that's one of its main points, I would say, to be lightweight, fast as hell and more responsive than any other synthesizer. Some phonemes, like the r, the z or the ch, for instance, could be a little less emphatically pronounced, they are clearly taken from other languages, but Czech is good in that it usually has just one phoneme for one written letter, so it actually sounds pretty understandable in that regard. Anyone remember the old Eureka A4 or Eureka Advanced, or even the Aria from the Robotron company? If you understood these well enough to use them daily, Czech Espeak sounds nice compared to that. :-)
Mine, as a now future poliglot, Have an interest to most impontantly, add some or update some of the languages that are constructive, like the now 4 of them (Esperanto, Interlingua, Lojban and Lingua-Franca-Nova), though more of them could be added like Interlingue (Occidental), which is kind of similar to Interlingua, and Novial, also similar in contects. I'm proposing to add the Occitan macro-language.
God bless you, a lot of mersy.
Gracias, thank you, your welcome, thumbs up, what ever, Luis Carlos. Have an awesome day!
Hungarian eSpeak. I suppose it's usable. With that said, it sounds terrible. for instance, try installing an app on a none english windows, and obviously your native eSpeak will read the installation screen in whatever lang the app is in. it's just terrifying. honestly, if I have to read my native, I either use vocalizer, and or slow down poor eloquence, and read native that way. which is interesting, as I usually have UK eSpeak on 35% with rate boost enabled. but when it comes to my native... around 20% at maximum, with rate boost enabled.
@6, Have you heard the polish Ivona speech synthesizers?
Does anyone know what the croak voice is in nvda. I still can't figure out that one. It sounds like the none voice.
The croak voice is essentially the default eSpeak voice with the flutter setting set to 20 (the flutter setting controls the croakiness in the voice).