Scribd
Upload
86 views

Speech Processing 15-492/18-492: Speech Recognition Template Matching

Template matching is a simple speech recognition technique that compares an input audio sample against stored templates. Dynamic time warping (DTW) allows the templates and samples to be of different durations by warping them for alignment. DTW works well for small vocabularies (<20 words) but larger vocabularies require extending the template model, such as stringing phoneme templates together. Reliability can be improved by averaging over multiple template examples and using distance metrics like Mahalanobis that account for variance.

Uploaded by

Shobhit Pradhan
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
86 views

Speech Processing 15-492/18-492: Speech Recognition Template Matching

Template matching is a simple speech recognition technique that compares an input audio sample against stored templates. Dynamic time warping (DTW) allows the templates and samples to be of different durations by warping them for alignment. DTW works well for small vocabularies (<20 words) but larger vocabularies require extending the template model, such as stringing phoneme templates together. Reliability can be improved by averaging over multiple template examples and using distance metrics like Mahalanobis that account for variance.

Uploaded by

Shobhit Pradhan
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 24

Speech Processing 15-492/18-492

Speech Recognition Template matching

Speech Recognition by Templates


A little history Matching Templates DTW (Dynamic Time Warping) Beyond template matching

Radio Rex (1922)


Toys always lead technology Call Rex and he comes out of his kennel

(Crystalradio.com and Rhys Jones)

Toy ASRTricks
Radio Rex
Recognizes vowel formants in EH

Voice activated toy train


Multilingual stop/go hashire/tomate

Toys pets dont need perfect ASR

Template Matching
Record templates from user
Store in library

Record ASR example


Compare against each library template

Select closest example For example


On a voice dialing system

Voice Dialing System


Library
Mom Dad Bob Marios Pizza

Lets Go Bus Information System

Matching in Time Domain


Duration
Will discriminate some examples But Mom, Bob and Dad will be confused

What about spectral properties

Matching in Frequency Domain

Mom

Bob

Different deliveries
We change durations
Two utterances are never the same

When it fails we change our delivery


Become more articular clearer

Dynamic Time Warping

Template

Sample Speech

DTW algorithm
Template
i i-1 j-1 j

Sample
For each square Dist(template[i],sample[j]) + smallest_of (Dist(template[i-1],sample[j]) Dist(template[i],sample[j-1]) Dist(template[i-1],sample[j-1]) Remember which choice your took (count path)

Multiple Templates
Compare against each Find closest Need to normalize scores
(divide by length of matches)

Matching Templates
Template Library Sample Word0 Word1 Word2

For Word in Templates Score = dtw(Template[Word], Sample); if (Score < BestScore) BestWord = Word; DoAction(Action[BestWord])

DTW issues
What happens with no-matches
Need to deal with none of the above

What happens with more templates


Harder to choose between Once variance greater than differences

Choose templates that are very different

DTW/Template Applications
Voice dialer Simple command and control Speaker ID

Speaker ID
Template Library Sample Speaker0 Speaker1 Speaker2

For Speaker in Templates Score = dtw(Template[Speaker], Sample); if (Score < BestScore) BestSpeaker = Speaker;

DTW
Advantages
Works well for small number of templates (<20) Language independent Speaker specific Easy to train (end user controls it)

Disadvantages
Limited number of templates Speaker specific Need actual training examples

More reliable matching


Distance metric
Euclidean

But some distances are bigger than others


Silence is pretty similar Fricatives are quite larger
A longer fricative might give large score A longer vowel might give smaller score

More reliable matching


Having multiple template examples
Individual matches or Average them together

DTW align all of the examples Collect statistics as a Gaussian


Mean and standard deviation for each coeff

More reliable distances


Instead of Euclidean distance
Doesnt care about the standard deviation

Use Mahalanobis distance


Care about means and standard deviation

Extending Template matching


String word templates together
Need to find word segmentation Word0 Word1 Word2

But there are many words

Extending template model


String phoneme templates together
A template model for each phoneme Sample k ae t Phoneme Templates Phone0 Phone1 Phone2

Summary
Speech Recognition by Templates
Good for simple small vocabulary tasks

Dynamic Time Warping (DTW)


Can match different durational examples

Averaging over multiple models Distance metrics


Euclidean vs Mahalanobis

You might also like