At the moment Kaldi is targeted more at people who are building ASR systems
than those who just want to use them. We may in future cater to needs such
as yours.
目前Kaldi面向構建ASR系統的人而不是使用ASR系統的人，以后可能會有API。
Regarding API, the idea is that you would write your decoder code in C++,
patterned after code already in Kaldi.
目前的做法是：仿照Kaldi中的已有的代碼，用C++寫自己的解碼器。
Dan
Kaldi的FastCGI，使得基於Kaldi的語音識別可以在支持FastCGI的HTTP服務器（如Apache、Nginx）。其中還包含一個基於HTML的客戶端，用於在網頁上測試Kaldi語音識別。
https://github.com/api-ai/asr-server