Microsoft's Acoustic Caller ID Patent

theodp writes "A new patent granted to Microsoft Tuesday for automatic identification of telephone callers based on voice characteristics covers constructing acoustic models for telephone callers by identifying words or subject matter commonly used by callers and capturing the acoustic properties of any utterance. Not only that, it's done 'without alerting the caller during the call that the caller is being identified,' boasts Microsoft in the patent claims."

Re:Wiretapping law

[VisceralLogic]

If it processes it in real time, it doesn't need to record it, really. Just pass through in and out.

Re:How is this different from existing voice recog

[Alien Being]

"How is what Microsoft claiming to do different from existing voice recognition systems?"

Existing voice recognition systems might be more acurately called speech recognition. They don't recognize the voice (who is speaking); they recognize the speech (what is being said). They can be categorized as speaker dependent or speaker independent.

Speaker dependent speech recognition (type 1) requires complex training by each user. It needs to know all the ways a person pronounces every possible phoneme. During use, it must be given the name of the speaker and a sound sample. It gives back the name of the phoneme. 2 inputs, 1 output.

Speaker independent speech recognition (type 2) is able to identify individual phonemes as spoken by a wide variety of speakers. 1 input, 1 output. That's what I would imagine is the important first step of what MS is claiming to do. Once a phoneme or two has been identified, the name of the phoneme and the captured sound sample can be fed to the type 1 algorithm and it would be able to output the name of the speaker.

Functionally it's different than existing "voice recognition" systems, but I seriously doubt it worthy of a patent.