rich_f Wrote:One thing I've noticed with Misaki-- she tends to say んが instead of plain old が. You can hear it more pronounced if you slow her down to about -7 or -8. (She talks pretty fast at 0.)

I haven't listened to the audio to hear what you are talking about first hand, but I do remember that in the past, one of my Japanese teachers has told me that が should actually be pronounced sort of nasally, like a slight ん　with it. That may only be in a certain dialect though.

Obviously, they're not going to be perfect. But they speak Japanese better than I do at the moment, and they sound pretty good. I find it's more useful for me to use to dump in words I'm not sure about, so I'm not just guessing at how a word is supposed to sound (creating bad input), so that alone is worth the money for me.

I don't *just* do this-- I also watch TV programs, I listen to podcasts, etc. I would NOT recommend doing any one thing in isolation, anyway.

BUT for uses where you want to know how a word "sounds," this is extremely useful. If you have Misaki already, and it works for you, and you can't afford the $35 for Show, then just stay with Misaki for now.

The Show voice sounds pretty much like what it sounds like on the website. It's a little less polished than Misaki, but it does have an easier voice range for me to follow along with. (I studied a lot of music growing up, so my voice training will kick in and I'll try to match Misaki's pitch out of instinct, so I wind up with my voice in weird registers.)

I was looking at the company's website, and they mentioned that the voices they sell vary in bit-rate sampling, (8, 11, and 16 kHz), as well as database footprint (from 16MB up to 500MB). From my experience, the Miyu and Show voices I bought were 16/500, so they were the same as the top of the line offered. Misaki is supposedly available in similar bit-rates and db sizes, so I don't know why there might be a difference between the server and your system, but it could be just about anything from sound cards to processor speed.

I have noticed that sometimes Misaki sounds smooth, and other times, she sounds a little rough around the edges. Who knows why. I'm not really too picky about that. I just find it easier to do it this way than to try to hunt down Japanese people at 11 at night when I just want to get a rough idea of how a word sounds.

The thing is, these voices have come so amazingly far compared to how they used to be, I'm not going to complain about it (too much.)

EDIT: If you're converting text to MP3-- then be forewarned. The quality of the MP3s TextAloud generates is horrid. The WAV files are perfect. If you're basing the quality of the voices on the MP3s, then I can totally see your point. I'd say save them as WAVs, and find a better program to use to rip them into MP3s. I don't create a lot of sound files myself-- I just dump text in the box and have it speak it out abit.

Well, I found a "trial" copy of Show (It is the 16/500 version) and I have to say, for some reason it sounds absolutely atrocious. We're talking Microsoft Sam bad. So, my question is, if anyone knows the answer: I am running TextAloud v2.228. If I upgrade to the latest, will the problem be fixed? Or is Show just horrible?

I have Microsoft Sam on my machine, and Show doesn't sound nearly that bad. Like I said, it could be software, it could be hardware. I have noticed that TA likes to eat a lot of CPU cycles. It also depends on your sound card/hardware, too. Bad hardware will produce bad sounds, and bad software will sabotage good hardware, too.

Can you link to a WAV file perhaps? A short one? Because it may be that it's not working, or it may be that you're too picky.

@nestor
That's not tinny... that's 7 kinds of awful. I'm not sure how they generate MP3s with that software, but whatever it is they're doing, they need to stop and find a new way to do it. I'm just going to stick to using it the way I'm already using it, and if I want audio to take with me, I'll just write WAVs and rip them to MP3 with proper software.

rich_f Wrote:That's not tinny... that's 7 kinds of awful.

Did you change the encoding options? I think it defaults to 32 kbps and 11.02 kHz, which is horrible. I changed it to 256 kbps and 44.10 kHz, and the mp3s it produces sound identical to both the wav files and speech generated within the TextAloud program. Well, I mean, it's probably not technically identical, since it's compressed, but you'd have to be a machine or some kind of freak of nature to tell the difference.

Well, apparently all I needed was a restart, because now Show sounds much, much better.

However, for some reason I can't get any voice to speak anything I type in TextAloud, nor can I create a wav or mp3 file.

[Katsuo]

taijuando Wrote:I have a Mac and a really slow PC in a closet somewhere. Is my only alternative to go back to the PC?

There's ドキュメントトーカ for Mac. I haven't tried it, but the web page samples don't sound too good.

@Mcjon
Yeah, I messed around with the encoding options, and all of the MP3s sounded bad. WAVs sound great, though.

@Ryuujin27
I don't know what to tell you-- I don't have problems using the usual JP typing mode in XP.

I understand that they're coming out with a new version 3 of the software sometime, but they're just saying, "It'll be done when it's done," for now.

Well, as it turns out my computer just doesn't like the software very much. It occasionally will let me use all the feature and occasionally won't. However, it always lets me use the firefox bar, so I can have Show read me news articles from goo.ne.jp.

So, guess I can't complain.

rich_f Wrote:@Mcjon
Yeah, I messed around with the encoding options, and all of the MP3s sounded bad. WAVs sound great, though.

@Ryuujin27
I don't know what to tell you-- I don't have problems using the usual JP typing mode in XP.

I understand that they're coming out with a new version 3 of the software sometime, but they're just saying, "It'll be done when it's done," for now.

Well, it's pushing annoyance, but I guess you can convert your WAVs into MP3s.

Can't wait to get back to Japan to get this program and try it out (my ship still "only" has a 7 meg pipe and NO WAY IN HELL of letting me download something 650 meg in size).

Well, it'd definitely be useful while you're on the ship, unless you have some native Japanese speakers in the crew.

I'm not too worried about converting wavs to mp3s. I don't have anything right now I want to listen to that badly, to be honest. I use it more for sounding out new vocab so I get the vowel length and pitch accent right.

Wow, let's all go and follow AJATT suggestions to the letter!

Using automatic Text-to-Speach software might make some sense in Chinese because the language works differently but for Japanese it's a sure way to:

(1) mess up one's readings as the software will struggle to recognise some character combinations (some people have already pointed this out)

(2) mess up one's sentence pitches

Most foreigners never acquire a decent pitch anyway but (1) should be a real concern...

Sorry to change the subject, but I was wondering if anyone could offer some advice about Production vs. Recognition. When adding new cards, from whatever source, how exactly do you decide whether to add the card as a Recognition or Production card (by which I mean going from kana to kanji)? Or are most people simply adding the fact and then testing themselves both ways?

Also, is there a simple way in Anki to see recognition cards first, before reviewing production cards (or vice versa)? I'm often on the train doing my reviews and can't very well whip out a pen and paper and start jotting down sentences, but doing recognition reviews is a cinch. If I could see those cards first, I could do the rest when I get home. Other times, I might want to plow through a bunch of production cards first before I head out.

nest0r Wrote:I'm only doing Production myself, though I imagine that if you're schedule engenders it or if you know a set of readings very well, you could just flip them. I think you can just edit the deck and toggle the Production/Recognition tags or whatever when you want to.

When you say you are only doing Production, what do you do with totally unknown kanji, such as those beyond the scope of RTK1 or RTK3? Are you learning how to write them now, or do you just include them on the Question side of the card and instead focus on producing the writing of those kanji you already know?

You don't have to write out sentences, you can just trace them with your finger in the palm of your hand... or on the head of the guy next to you.

The whole idea isn't to kill 8,000 trees, but to be able to go from kana->kanji.

Picking sentences to go kana->kanji is pretty easy for me-- I only pick ones where I'm learning new kanji readings. If I already know the readings (like in the case of grammar sentences), then I go one way only.

@Serge-- Ghinzdra has double-checked the Misaki voice's abilities with his Japanese teachers. He wrote that up a few pages back in the thread. The voices are not perfect, and they're no replacement for the Real Thing, but for those of us who don't have a native speaker nearby, they do well in a pinch to get pronunciation down. If you haven't had a chance to play with one, you really should check out the demo. They've come a long way in the last few years.

Like any other tool, you use it when you need to. No need to use it all the time, just use it when you have trouble pronouncing something, and don't have a Japanese person handy. I find it's pretty handy for sounding things out. I use it maybe 10-20 times a review session, which is about 100 cards. I just dump text into the reader-- I don't mess with making sound files over and over.

And if you're paying attention, you can tell when the software changes a reading pretty easily. A lot of times, I find simply inserting a comma between some of the offenders will straighten things out.

I currently go both production and recognition on my cards. Production takes the longest due to having to write out the sentences. Recognition is just reading, so no big whoop there. If you use anki, this should be no chore at all so long as you kept your kana sentences to one fact entry.

For Kanji that's not RTK1, I've been adding them into Anki as they show up in sentences. I've been making a habit of converting all the sentences I do into kanji. So far out of 650 sentences only about 15 kanji are RTK3 (wolf, dust, ornate, goro, etc). These too I'm doing production and recognition.

Serqe, for using the voice recognition, it may be going solely voice to kanji is the ultimate way to practice. If you recognize what is being said, then recognizing it in kana form should be no problem. If you can write what you hear, then saying what is written should not be a problem. It makes complete sense to me, and is something I'll be attempting when I return to Japan in a couple of weeks. Hell, just going kana to kanji (something I've only done for two months) after a post on AJATT about it has GREATLY increased my On yomi, Kun yomi and vocabulary.

Honestly, I would say take both the Production and Recognition cards and use both. It will only help you, really.

Some say that doing it one way will take care of the other on it's own, but why wait? I say do both. It definitely won't hurt.

I think that both are essential. Khatzumoto himself seems to think that doing production is enough, but I don't agree on this.
I've tried it for some time, and doing production (kana->kanji) is great for your ability to recall kanji while writing. But it seems to me that there is a different process kicking in when doing recognition. Like some sort of visual automatism that is responsible for your lighting fast recognition of words as a whole. When reading one doesn't look at isolated characters, after all.
So my opinion would be that both ways are necessary if one wants to improve in reading AND writing fluently.

nest0r Wrote:Hmm. You should go back and read about customizing the TTS if you're having difficulty with readings. I recommend only using sentences that you have kana readings for. Also, try Misaki if you have issues with the voices you're using, as her sentences sound very natural, though you shouldn't just listen to the TTS for that sort of thing, as the AJATT page suggests, make sure to get plenty of listening and samples from elsewhere.. Don't be too literal in interpreting the AJATT method Serge, make sure you pay attention to the oblique suggestions, and also, try to find your own way. Good luck!

rich_f Wrote:@Serge-- Ghinzdra has double-checked the Misaki voice's abilities with his Japanese teachers. He wrote that up a few pages back in the thread. The voices are not perfect, and they're no replacement for the Real Thing, but for those of us who don't have a native speaker nearby, they do well in a pinch to get pronunciation down. If you haven't had a chance to play with one, you really should check out the demo. They've come a long way in the last few years.

Like any other tool, you use it when you need to. No need to use it all the time, just use it when you have trouble pronouncing something, and don't have a Japanese person handy. I find it's pretty handy for sounding things out. I use it maybe 10-20 times a review session, which is about 100 cards. I just dump text into the reader-- I don't mess with making sound files over and over.

And if you're paying attention, you can tell when the software changes a reading pretty easily. A lot of times, I find simply inserting a comma between some of the offenders will straighten things out.

Thank God I'm past the stage where I need all of that... and even if that wasn't the case I would think twice before allowing TTS software to teach me a language. And why would anyone have 'trouble pronouncing something' in such a phonetically straightforward language as Japanese is really beyond my understanding.

But by all means - please feel free to go ahead and mess up your sentences pitches...

nest0r Wrote:Oh! Congratulations then, senpai! It must have taken you a while. But it sounds like you're still having trouble with sentence pitches and the like. Even though Japanese is relatively straightforward with the consonants and vowels and segments and suchlike, it's important to get the 'flow' correct. I suggest you get more listening and speaking practice with native speakers and non-TTS sources, as well as checking out this thread and the sections about newer TTS voices, such as this one called Misaki by Neospeech. You can even edit the way they speak, though it's seldom necessary. Well, that covers your concerns. Remember, have fun!

What a charming way to promote TTS software... :-) One would almost think that you're pretending to misunderstand my remarks with that sole purpose in mind... :-)

nest0r Wrote:hehe, If you've realized what I'm doing, then you're intelligent enough to know why, ie what about your remarks inspired me to do so, and that's enough for me.

I happen to like the friendly spirit of this forum so I'd rather not take any guesses at your motives... :-) Always happy to inspire, though... :-)

Whether or not you think it's good for someone's pronunciation, it certainly couldn't hurt to use it as a form of dictation for flash cards.

Speaking of the AJATT method, after following "most" of the posts in this thread, I still
can't seem to get a grasp of the method by which the flash cards are created. I can't seem to find a proper way of creating flash cards having sentences in Anki, since I can't seem to find a method that suits me. So any suggestions might help. By method I mean, how does one create the cards in Anki? For example -

Card (production)

Q - Kana, English Translation, Audio
A - Kanji Sentence
Extras - Vocabulary Definition (Japanese or English)

Card (recognition)

Q - Kanji Sentence
A - Kana Reading, English Translation, Audio
Extras - Vocabulary Definition (Japanese or English)

And of course, by using the method mentioned above, one can drill vocabulary, and kanji and maybe pronunciation, but what about grammar? Will it be good enough just to highlight the important grammar points per card without any explanation for it. Or is it advisable to add grammar notes ? Won't that be just overloading the card with too much information ?

Sorry for asking too many questions. ^_^;;

atreya Wrote:Sorry for asking too many questions. ^_^;;

These are good questions and I would like to thank you for asking them, it saves me the time I would have needed to write them down myself .

Everyone has a different approach to doing these card sets. A lot of what you do varies based on what you're comfortable with. If you need a lot of hand-holding, then you'll be more likely to cram more data on a card. If you don't, you won't. Depends on you.

What I do:

Production: Sentence full of kanji. Whatever I want to highlight may be underlined or bolded. May not.

Translation: English translation of the sentence, followed by readings of kanji I don't know. Like this: 予定　　よてい. I don't put a translation in, because I should be able to guess it from the sentence. (Haven't had problems with this.) If I need a translation, I'll put it in.

Reading: Reading of the sentence in kana.

I always create a card the goes Recognition-> Translation + Reading. When I'm learning new kanji, I'll create card for Reading-> Recognition + Translation.

When I'm learning grammar, I'll make a note of grammar points in whichever language works best on the Translation part of the card. I usually only go one way with those cards... but sometimes I'm lazy and I create 2 anyway.

Experiment. See what works for you. There's no 'right' or 'wrong' way to do it.

I do find that going from kana->kanji helps me write out kanji a lot, and going from kanji->kana helps with reading. I haven't messed with dictation, to be honest.