Recognition by comparison with pre-recorded words

问题:

Requirement:

I want a system/algorithm that is similar to the call system in the car that you say a name that you want to call, it will find out that name match your voice. My requirement is that I will use pre-recorded voices instead of Name.

e.g.: User record target voices for several words in non-English language, e.g.:

• "Keyboard-JP-01.wav" => Keyboard
• "Monitor-JP-01.wav" => Monitor

After some training or even no training, when the user speak a word again, the speech recognizer should able to identify which word it is.

Known tools:

I searched existing speech recognition algorithm, but cannot find one, most of them are recognizing existing languages (kind of speech to text).

It also different from phonetic algorithm which seems only for English.

Thanks

回答1:

Dynamic Time Warping is a traditional and fairly simple algorithm for this kind of speech recognition, though I'm sure there must be better methods nowadays.