You are on page 1of 1

Home Automation(Speech Recognition)

Algorithm:
Get sound from a mic Divide it into smaller chunks Convert Pulse Code Modulation digital audio(from mic) to frequency domain by taking fourier transform Divide the audio segment into phonemes(alphabets in words spoken) Train a data set by using phonemes for pattern matching Compare run time phonemes with dataset Arrange phonemes to make detected words If failed Neural Nets might be utilized

Problems
Speech must not be in a continuous manner Higher processor resources required Phonemes with different variations are difficult to recognize

Phonemes Recognition:
Variations of all phonemes will be indexed into a queue Probabilities of each variation of phoneme are calculated manually These probabilities are used during testing time Different probability models can be used which will vary depending upon performance and computation intensiveness Hidden Markov Models are required for differentiating different phonemes i.e. their beginning and end Depending upon the probability of each phoneme and its variant one variant of a phoneme can be selected with the highest probability and other variants are pruned In differentiating phonemes by markov model, the selection of phonemes will be done on the basis of silence phase in-between each phoneme.

Noise Reduction:
For Noise deduction Backgrounds voices must be minimized Discrete instead of continuous speech must be used A threshold level must be marked for Discrete speech detection

You might also like