Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Wakeup words support for Chinese #1

Open
juncaofish opened this issue Nov 28, 2016 · 13 comments
Open

Wakeup words support for Chinese #1

juncaofish opened this issue Nov 28, 2016 · 13 comments

Comments

@juncaofish
Copy link

Hello,
Just need a local support feature for voice wakeup~
I notice this file respeaker/pocketsphinx-data/dictionary.txt .
So my question is if I want to set a customised wakeup word in Chinese, like "魔镜", can I add to the dictionary something like 'mojing' M UE J IN ?
Thanks.

@xiongyihui
Copy link
Contributor

To recognize Chinese words, you need another acoustic model, for example
https://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/Mandarin/.

But the mandarin acoustic model is not good enough. You may need some training for the model.

@juncaofish
Copy link
Author

Thanks for replying.
So is MT7688 capable enough to run the mandarin acoustic model?

@xiongyihui
Copy link
Contributor

The mandarin acoustic model is too big for MT7688 to do continuous speech recognizer. You need a small one.

@juncaofish
Copy link
Author

I find a tool https://github.com/Kitt-AI/snowboy which seems to be a perfect solution for Chinese hotword local detection. However, it seems it cannot be targeted to MT7688 either for the hardware architecture limit. Do you have any suggestions on how to take less effort for Chinese hotword detection task? Thanks very much.

@xiongyihui
Copy link
Contributor

It seems the snowboy is the most accurate. Ask for kitt-ai to compile a MIPS version of snowboy maybe work. There are old reversions of mandrarin acoustic model in cmusphinx repo which is smaller and should be able to run on MT7688, or train a smaller model by hand.

@juncaofish
Copy link
Author

I'm afraid the MT7688 is not a supported target for snowboy. Take a reference at discussion here: https://groups.google.com/a/kitt.ai/forum/#!topic/snowboy-discussion/iL4rRZRUk4U

@juncaofish
Copy link
Author

juncaofish commented Dec 1, 2016

is it possible to use Xunfei wakeup on Respeaker, I mean compile the sdk for 7688's mips platform? It seems like a good solution if cost is not considered as a cons.

@xiongyihui
Copy link
Contributor

I applied a sdk of xunfei's stt and stt running on 7688, it worded. The wakeup sdk should work too.

@juncaofish
Copy link
Author

juncaofish commented Dec 1, 2016

Since I have tried the online sdk on Raspberry Pi, I guess you mean the online tts/stt service of xunfei. Do you need to submit the compile chain to their engineer? I remember the sdk only provides .so for x86/x64 .

@xiongyihui
Copy link
Contributor

Yes, for c sdk for 7688, we need to submit the toolchain.

@calvinwongsc
Copy link

I am intersted in using xunfei's wakeup sdk.... i have download the pack from there, and how can i use it in respeaker?

@jiaoxu
Copy link

jiaoxu commented Aug 3, 2017

yihui 你好
我在mac上执行bing_translator.py的时候,提示DEBUG:urllib3.connectionpool:https://oxford-speech.cloudapp.net:443 "POST /token/issueToken HTTP/1.1" 401 143
00000000000recognition connection failed,我已经提供了申请的密钥 1: ,这个错误是什么问题,国内网络限制吗?

@solidssss
Copy link

Is there any way to actually run snowboy detect on 7688, if so how? Thanks

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants