Your Features in Text to Speech Software

[morriss003]

What would you add to a Text to Speech software package?

I've been googling TTS software, and I am amazed at how primitive it still is. I would have thought that I could open an rtf file, highlight a section of dialogue and assign it to a voice, of which I thought there would be many.
Apparently not.

[DuaneAA]

I use TextAloud software( http://www.nextup.com/TextAloud/ ) and it will do the 'highlight a section of text and change the voice' like you are talking about. However I almost never do it for a couple of reasons.

First, if you are going to do that for a novel length document, it would take a lot of time to go through it line-by-line and implement this for every different character. I use this software mostly to proofread things I write, as I can 'hear' lots of typos that I don't see when I am reading it. And frankly, I like to listen to some of my stories while I am in the car to help motivate myself to keep working on them.

Second, I have occasionally used this TTS software to convert fanfiction or ebooks I have found and bought in RTF, which this software can convert. But going through a whole book written by someone else to set up the proper voices sort of defeats the point, as you will have now read the whole book in setting up the TTS version with distinct voices.

Third, the best, most natural sounding voices are sold as separate files and are rather expensive running $30-$50 each. It is difficult to justify that kind of money to have 10+ distinct voices available.

I have found one British female voice I really like (Rachel22 by Acapela) and I use it for everything. I have literally spent hundreds of hours listening to this voice read things to me and I am not tired of it yet. Perhaps once a year I troll through the voices that are currently available and I haven't found one I like better. If you would like to hear a sample of this voice, I have posted a link to the first chapter of one of my books at the bottom of this blog post. The sample runs 14 minutes: http://duaneaakre.wordpress.com/2012...-a-real-novel/

Hope this helps,
Duane

[morriss003]

Originally Posted by DuaneAA

I use TextAloud software( http://www.nextup.com/TextAloud/ ) and it will do the 'highlight a section of text and change the voice' like you are talking about. However I almost never do it for a couple of reasons.

First, if you are going to do that for a novel length document, it would take a lot of time to go through it line-by-line and implement this for every different character. I use this software mostly to proofread things I write, as I can 'hear' lots of typos that I don't see when I am reading it. And frankly, I like to listen to some of my stories while I am in the car to help motivate myself to keep working on them.

Second, I have occasionally used this TTS software to convert fanfiction or ebooks I have found and bought in RTF, which this software can convert. But going through a whole book written by someone else to set up the proper voices sort of defeats the point, as you will have now read the whole book in setting up the TTS version with distinct voices.

Third, the best, most natural sounding voices are sold as separate files and are rather expensive running $30-$50 each. It is difficult to justify that kind of money to have 10+ distinct voices available.

I have found one British female voice I really like (Rachel22 by Acapela) and I use it for everything. I have literally spent hundreds of hours listening to this voice read things to me and I am not tired of it yet. Perhaps once a year I troll through the voices that are currently available and I haven't found one I like better. If you would like to hear a sample of this voice, I have posted a link to the first chapter of one of my books at the bottom of this blog post. The sample runs 14 minutes: http://duaneaakre.wordpress.com/2012...-a-real-novel/

Hope this helps,
Duane

Thanks for the tips Duane. Highlighting dialogue in my own work is what I was looking for. Why do you think voices are so expensive? I would have thought that software would be available to create voices.

[DuaneAA]

I am no expert on what goes into making a 'voice'. I do know the file for a good voice is typically about 500 Mb. So I assume it must be a difficult, challenging task and why I have never seen software to create your own voices.

Since English has something over 100,000 words and I don't know if that includes names for people or places, it is amazing how well it does. I have even made up words in imaginary languages (including letters 'borrowed' from other languages like German and Spanish) and as long as you have a reasonable mix of vowels and consonants; it does a remarkably good job of pronouncing them without a hesitation.

The only place the software ever has problems is with common words that have several different pronunciations. It seems like it screws them up a lot, but I'm sure it just you only notice when it uses the wrong one. Among the most annoying ones are 'minute' (it seems to prefer the pronunciation for a small quantity when 98% of the time you really want the word referring to a unit of time), read (it defaults to the present tense, when I usually want the past tense), and it annoying spells out h m m rather than saying hmm. Some of these you can fix by tweaking the wording (like typing 'red' when you want the past tense of 'read') and some you can fix with their learning function (that I have always been to lazy to learn). But again it all just takes time, if you are trying to tweak something that is hundreds of pages long. So I have just learned to live with a small percentage of mispronunciations.

I have been using this software for 4-5 years, but I really don't know how it works other than it can't be a simple look up table, as the pronunciation of words will change to some extent with context. In particular, if the sentence ends with a question mark, the voice changes the inflection so you hear the sentence is a question.

Duane

[morriss003]

Originally Posted by DuaneAA

The only place the software ever has problems is with common words that have several different pronunciations. It seems like it screws them up a lot, but I'm sure it just you only notice when it uses the wrong one. Among the most annoying ones are 'minute' (it seems to prefer the pronunciation for a small quantity when 98% of the time you really want the word referring to a unit of time), read (it defaults to the present tense, when I usually want the past tense), and it annoying spells out h m m rather than saying hmm. Some of these you can fix by tweaking the wording (like typing 'red' when you want the past tense of 'read') and some you can fix with their learning function (that I have always been to lazy to learn). But again it all just takes time, if you are trying to tweak something that is hundreds of pages long. So I have just learned to live with a small percentage of mispronunciations.

Duane

I appreciate the info. Have you ever seen a way to tell the software what definition it should use for a particular word? For example, "minute." If you could highlight the word and then click on a definition, the software would know how to pronounce the word.

Clearly this type of app has a long way to go, but I suspect that there will be a lot more work on this type of app real soon. If I understand correctly, Siri is actually in the cloud and constantly learns from everyone's usage.

[DuaneAA]

Originally Posted by morriss003

I appreciate the info. Have you ever seen a way to tell the software what definition it should use for a particular word? For example, "minute." If you could highlight the word and then click on a definition, the software would know how to pronounce the word.

I believe this software will do that. I have just been too lazy to figure it out.

Originally Posted by morriss003

Clearly this type of app has a long way to go, but I suspect that there will be a lot more work on this type of app real soon. If I understand correctly, Siri is actually in the cloud and constantly learns from everyone's usage.

Do real people actually talk to Siri? I like the basic concept, or at least where it is headed as some kind of personal assistant. But I think for the present I would be happier to interface with it by typing than talking to it. It just looks so . . . awkward . . . from what I have seen in the commercials, particularly if you are out in public (Of course, I haven't seen any of the celebrities in the commercials using it anywhere except in the privacy of their home, almost as though even Apple knows how embarassing it is). However, I guess they have to start somewhere.

Duane

[twowheels]

Originally Posted by DuaneAA

Do real people actually talk to Siri?

I use it a LOT!

While driving, I just touch the button on my bluetooth headset and say "read it to me" after a new text message comes in. I can then just say "reply" and then dictate the message. It's about 98% accurate (not measured) and even manages to get the subtleties right. I then ask it to read it back to me before telling it to send. Completely eyes free, and only requires a tap of the button next to my ear to initiate.

Before bed I take my phone out of my pocket, hold the home button, say "wake me at 8 am" and plug the phone in and set it down. It reports to me that it has set the alarm for 8, and I'm done. MUCH faster than unlocking, finding the clock app, changing to the alarms tab, choosing edit, choosing the alarm to edit, changing the time, then confirming it.

To call stores, I just say "call Target" and it does...


In public I use it by speaking in a low whisper and then putting it up to my ear before the answer comes back, which causes it to answer via the earpiece rather than the speaker. There's an option, which I haven't tried, to have it just automatically turn on Siri when you put the phone to your ear, so it would be completely silent to people nearby.

For long e-mails I find it MUCH faster to dictate the message than to type them on the on-screen keyboard (though a real keyboard is still faster). You have to learn to speak differently, but it works well once you get the hang of it.... For example: Hey Dave exclamation point newline how are you doing question mark newline will you be able to meet me at 4 questionmark the openquote boys closedquote openparens well comma not really boys closed parens are looking forward to seeing you period newline later comma newline Billy

(my name's not Billy :-) )



BTW, your TTS sample just about drove me nuts! Waaaaaayyyy too slow for me! :-)

[twowheels]

Just to demonstrate, I am "writing" this entire response using the voice recognition on my iPhone 4S:

Hey Dave!
How are you doing?
Will you be able to meet me at four? The "boys" (well, not really boys) are looking forward to seeing you.
Later,
Billy


The only real problem that I have, is that it ignores duplicate
new lines, so I can't insert extra blank lines. I had to do those manually.

[DuaneAA]

Originally Posted by twowheels

I use it a LOT!

That's interesting to hear. I have a few friends with iPhones, but I have never asked if they use Siri.

I have never had an iPhone, my last several have been androids and before that I had several Windows phones before most people had ever heard of smartphones. Currently, I have a Galaxy Note and I love the large 5" screen. It is great for reading books and watching videos at the gym.

Shortly after Siri came out, someone hacked together a quick android rip-off called Iris (Siri backwards). You could ask it wiki type questions, but that was about the limit of its functionality. I played with it for about 45 minutes before the novelty wore out. It sounds like Siri has a lot more utility.

For long e-mails I find it MUCH faster to dictate the message than to type them on the on-screen keyboard (though a real keyboard is still faster). You have to learn to speak differently, but it works well once you get the hang of it.... For example: Hey Dave exclamation point newline how are you doing question mark newline will you be able to meet me at 4 questionmark the openquote boys closedquote openparens well comma not really boys closed parens are looking forward to seeing you period newline later comma newline Billy

(my name's not Billy :-) )

This thread started out about TTS, but it is interesting you have brought up the exact opposite - voice recognition. It's been a number of years ago - at least five or six - but several times I explored the voice recognition approach to writing by using the Dragon Naturally Speaking software. I think my brain just isn't wired that way. Having to speak the punctuation drove me crazy and kept breaking my train of thought. And I spent more time going back to fix the punctuation and other typos, than it would take to simply type things. I think for anything more than a sentence or two, it is always going to be a keyboard for me.

BTW, your TTS sample just about drove me nuts! Waaaaaayyyy too slow for me! :-)

The software I use allows you to adjust the speaking speed. I do have it set several notches slower than the default. I have tried several different speeds and that happens to be the one I personally like the best.

Duane

[twowheels]

A real keyboard, sure. I type well over 100wpm, so of course. On screen keyboard, though and I definitely prefer to speak.

BTW, android can do most (all?) of what I described, it's just not marketed as much. I did much of the same stuff on my old Motorola Droid before the iPhone.