Speech Recognition
-
Try CMUSphinx: http://cmusphinx.sourceforge.net/wiki/ They have an open-source C library, their documentation looks pretty good, and they have a Hello World speech recognition tutorial.
Keep in mind that you need to create (or download) a language model and an acoustic model. They only provide an acoustic model for American voices, which may produce inaccurate text for non-American speakers. You'll probably also need an understanding of statistics and linguistics.
-
That's the definition of "Speech Recognition": Converting spoken words into text. :)
-
I found one that supports several languages, a library seems to be good for us to discuss, what do you think?
Languages:
en-us American English
en-sc English with a Scottish accent.
af Afrikaans
bs Bosnian
ca Catalan
cs Czech
de German
el Greek
eo Esperanto
es Spanish
es-la Spanish - Latin America
fr French
hr Croatian
hu Hungarian
it Italian
kn Kannada
ku Kurdish
lv Latvian
nl Dutch
pl Polish
pt Portuguese (Brazil)
pt-pt Portuguese (European)
ro Romanian
sk Slovak
sr Serbian
sv Swedish
sw Swahihi
ta Tamil
tr Turkish
zh Mandarin Chinese
cy Welsh
grc Ancient Greek
hi Hindi
hy Armenian
id Indonesian
is Icelandic
jbo Lojban
ka Georgian
la Latin
mk Macedonian
no Norwegian
ru Russian
sq Albanian
vi Vietnamese
zh-yue Cantonese Chinese
-
eSpeak is a Speech SYNTHESIS program.
You wanted a Speech RECOGNITION library, right?
Speech Synthesis = Text-to-Speech
Speech Recognition = Speech-to-Text -
Which language(s) do you want?
Google can help you
-
[quote author="l3e0wulf" date="1350170935"]I want almost all languages, if possible. (Like eSpeak, that supports many languages)[/quote]That's just not possible. Even though CMUSphinx supports base models for several languages, many developers still choose to "train" their software with their own audio recordings, to improve accuracy.
Understand that speech recognition is extremely complex. You need to support more than just the vocabulary: You also need to support the speaking style. Think of England, Scotland, USA, Australia, New Zealand, South Africa, Brazil, Germany, Singapore, India -- people in these countries can speak English, but they sound very different. Speech recognition software designed for one country will be very inaccurate in another country. And that's only for one language!
So, when you develop speech recognition software, you need to "train" it to match your users. I think it's too expensive and time-consuming for library-makers to support a large variety of languages and styles -- their priority is to spend time on designing algorithms, not on collecting speech recordings.
What are you plans for your software, and which users do you want to target?
-
Can some one please let me know whether QT5 has support for speech command ?
Regards,
Souvik -
Hello there,
I've been looking lately for a complete toolkit for developing a research project which we have to use "Speech Recognition" and "Speech Synthesis". The main library and API should have "Speech Recognition" in C++ with an integrated "acoustic model", "linguist model" and of course with a "decoder".
During this project I'll have to use both techniques, "Speech recognition" and "Speech Synthesis". Both in* Portuguese (Brazilian)* and English.
I plan using the native Qt library for Speech Synthesis:
"Qt TTS":https://gitorious.org/qt-speech/qt-speech/source/6ddec1ee6dbcaa4ca74a625bdc87c8ace08bb045:But I know the Speech recognition process is way harder, I tried using CMU sphinx ("CMU Sphinx":http://cmusphinx.sourceforge.net/wiki/tutorialpocketsphinx) but didn't find examples and concrete tutorial for C++ developing. Also tried VOCE ("VOCE Sourceforge":http://voce.sourceforge.net/) but it can't be handled for portuguese.
I'm an open mind guy, if you are developing something like this please contact me so we can share some stuffs (vitorsgobbi@hotmail.com)
-
Hi,
After googling a lot and studying about "speech recognition" I realized CMU Sphinx is the best option for me.
For those who needs to train an acoustic model here is the tutorial.
However, it is complex and requires lots of work and time though."Acoustic model":http://cmusphinx.sourceforge.net/wiki/tutorialam
This is a hello world example in .C
"Hello world sample":http://cmusphinx.sourceforge.net/wiki/tutorialpocketsphinx
And this is for those that are Brazilians developing their own application in speech recognition like me. You can find a Portuguese acoustic model and the language model here:
"LAPS UFPA":http://www.laps.ufpa.br/falabrasil/downloads.php
