Recognition of individual votes
I am planning to write a conversation analysis software that will recognize individual speakers, their level and intensity. Pitch and Intensity is somewhat simple (step through autocorrelation).
How do I recognize individual speakers so that I can record his / her functions? Would it be sufficient to store the heuristic for each speaker frequency? I can assume that only one person speaks at a time (not strictly overlapping). I can also assume that, for training purposes, each speaker can record the minute value of the data before the actual analysis.
a source to share
Step and intensity alone don't say anything. You really need to analyze how the pitch changes. To identify different speakers, you need to convert the speech sound to some kind feature space
and then compare with your speaker database in that function space. The general term you want to use for Google is prosody
- see for example http://en.wikipedia.org/wiki/Prosody_(linguistics) . While you are Googling you can also read up on speaker identification
aka speaker recognition
see for example http://en.wikipedia.org/wiki/Speaker_identification
a source to share
If you are still working on this ... are you using speech recognition on audio input? Because Microsoft SAPI, for example, provides an application with a rich API for digging into the speech sound wave, which can make the speech recognition problem more acceptable. I think you can get phonemes in waveform. This will allow you, for example, to perform a vowel spectrum analysis that can be used to generate functions to distinguish between loudspeakers. (Before anyone starts mumbling about pitch and volume, keep in mind that the formant curves come from the shape of the vocal tract and are fairly independent of the tone, which is the frequency of the vocal cord, and the relative position and relative amplitude of the formants (relative!), Regardless of the total.) Phoneme duration in context can also be a useful feature.The distribution of energy during "n" sounds could provide a "nasal" function. Etc. Just a thought. I expect to work in this area on my own.
a source to share