We are looking to create an iPhone app that simply counts the number of words spoken from an audio stream. For example, user hits “start recording”, says a sentence like “Hi, I am Mike” and then hits “stop recording”. The output would be 4 words. We don’t care about what words specifically are spoken, just the number. Is this possible with OpenEars?

Thanks!

This would be possible to do if the words were identifiable (i.e. you knew them in advance). Without knowing the words which are spoken, you can’t really count number of words, because you don’t know what comprises a word versus some arbitrary combination of syllables.

Two thoughts:

Is there any functionality to measure syllable nuclei:
https://sites.google.com/site/speechrate/Home/how-does-the-script-work

OR

Can you simply do a speech-to-text and then just do a total count on all of the words?

Is there any functionality to measure syllable nuclei:
https://sites.google.com/site/speechrate/Home/how-does-the-script-work

No, but this wouldn’t tell you how many words the user had said, just how many syllables. You can currently get an middling-good attempt at phoneme transcription out of Rejecto, but again, knowing parts of words doesn’t tell you anything about how many words were said unless you already know the words, in which case you should just create a language model or grammar using those words and then count utterances of them directly.

Can you simply do a speech-to-text and then just do a total count on all of the words?

Yes, I think I would do it this way using a custom language model, but it would require knowing the needed vocabulary in advance.