High-Quality Text-To-Speech engine for personal us

问题:

I'm looking for a high-quality TTS engine that I can afford (let's say less than 1000$). So far, I've tried flite and festival with default voices. However, while the results are certainly understandable, technical texts are hard to follow.

Commercial TTS solutions from Loquendo and Readspeaker sound way better. However, these companies don't seem to be willing to sell their product to mere mortals - I can't find a price on either's homepage.

So, what are good TTS solutions for personal use?

回答1:

Purchase it from NextUp.com site:

NextUp.com sells the best, most natural-sounding Text to Speech voices with more than 20 languages and many accents available.

• "Natural Voices" from AT&T ($35.00)
• "Verbose" from NCH ($29.99)
• "TextAloud" from NextUp ($29.95)
• "RealSpeak" from Nuance/ScanSoft ($45.00)
• "Acapela" from Acapela Group ($35.00)
• "Cepstral" from Cepstral LLC ($29.99)

There are also:

• "CrazyTalk6 PRO" from Reallusion ($149.95)
• "IVONA Voices" from ivona ($45)
• "Dragon Premium" from Nuance ($199.99)
• "VoiceText" from NeoSpeech (about $74.95)
• "FonixTalk" from Fonix Speech (about $45.00)
• "NaturalReader" from NaturalSoft ($199.50)
• "Cerevoice" from Cereproc (about $36)

回答2:

I was always impressed by Ivona http://www.ivona.com/?set_lang=en They have a cheap personal version called expressivo (add .com - can't post more links) which is just $45. I know people who watch movies with Expressivo reading the subtitles, so it's actually very very good.

回答3:

You can download better quality voices for festival than the ones shipped with it:

These seem to be the ones with the highest quality right now:
HMM-based Speech Synthesis System (HTS)

These seem to be ok too:
The MBROLA Project

Source: HOWTO: Make festival TTS use better voices (MBROLA / CMU / HTS)

回答4:

AT&T has a product called Natural Voices. I think that is sounds amazing compared with all of the other products out there. I'm not sure about pricing though.

http://www2.research.att.com/~ttsweb/tts/demo.php

回答5:

There are also some quite impressive open source solutions.

This one sounds quite impressive. http://freetts.sourceforge.net/docs/index.php

回答6:

Please check this site

Basically, it's a high level tutorial to use voices available from Android on Linux. It's quite general, but the technic should work for most TTS engine.

回答7:

I did some research on the topic in 2007 and tried several text to speech systems to read articles or convert them to mp3. I am surprised how little progress the consumer TTS products have made since then.

First I bought TextAloud by NextUp because in the web samples the voices sounded natural. However, it turned out that the way parts of the sentences were stressed made it really hard to understand scientific texts. I don't know if that has improved.

I then found the VoiceReader Home by Linguatec (49 Euro per voice) which did a very good job and I haven't found a better solution for myself since then. Linguatec just did an update which I have been using for a few weeks now and the quality improved even more.

However, I don't like their GUI and integration as much. I basically copy all texts that I want to read into the GUI window. Yet, the new version can also read pdf and word documents from file. And for me the most important factor is still the ease of listening and understanding of text even with a complex structure.

Linguatec looks like a small German company. I don’t know if they have their own TTS engine or use an external one.

I am not associated with Linguatec in any way and would be very interested in alternative suggestions!

回答8:

Surprisingly, loquendo does sell stuff: Price list for Pay as you go TTS. Unfortunately, this works online when online, and costs a fortune for on-demand voice generation (30 minutes a day would come just under 120.000€ per year).

回答9:

Nuance vocalizer but im afraid they do not say something about the prices. Nuance has a lot of high quality speech applications, so maybe they can make you happy.

回答10:

I've used AT&T Natural Voices, they seem to sound most human to me.

回答11:

For personal use, I'm using VocaTalk Personal Podcast. The app enhances speech, makes it stereo, puts bg music, generates mp3, uploads to skydrive/google drive, syncs up with iTunes or zune and what not. Perfect for on-the-go listening. http://www.vocamedia.com/

回答12:

For personal use you could also use "TextAid" from ReadSpeaker. It is web based personal reader and uses Acapela voices. It also has a translation feature! http://www.readspeaker.com/readspeaker-textaid/

回答13:

What about generating speech online? You can use the free website audiotext.ws text to speech to convert English texts to speech.

回答14:

If you are looking for indian accent based output, take a look http://www.indiantts.com

It seem to sound most human ,Indian dialects.They offer download mp3 option I think from text.