8/6/2019 T1 Intro Speech Processing

1/21

2. Introduction toSpeech Processing

8/6/2019 T1 Intro Speech Processing

2/21

The Speech processing stack

Speech Applications: Coding, synthesis, recognition,understanding, speaker verification, language translation, speed-up/slow-down

Speech Measurements: energy, zero crossings, autocorrelations

Speech properties: speech-silence, voiced-unvoiced, pitch,formants

Speech representations: temporal, spectral, homomorphic, LPC

Fundamentals: acoustics, linguistics, pragmatics, speechperception

8/6/2019 T1 Intro Speech Processing

3/21

SPEECH GENERATION ANDTRANSMISSION

8/6/2019 T1 Intro Speech Processing

4/21

Tractament Digital de la Parla 4

Speech Chain

SPEAKER

HEARER

(Denes and Pinson)

8/6/2019 T1 Intro Speech Processing

5/21

Tractament Digital de la Parla 5

Speech Production/Perception

(after Flanagan)

8/6/2019 T1 Intro Speech Processing

6/21

Tractament Digital de la Parla 6

Speech Processing Diagram

8/6/2019 T1 Intro Speech Processing

7/21

Application of digital speechprocessing

Speech coding

Speech Synthesis (from text to speech)

Speech recognition

Speaker/language recognition Many others

8/6/2019 T1 Intro Speech Processing

8/21

SPEECH CODING

8/6/2019 T1 Intro Speech Processing

9/21

Tractament Digital de la Parla 9

Speech Coding The aim of speech coding is to compress (and then

decompress) the speech waveform without any loss oflistenabilityor intelligibility.

Various standards exist for speech coding.

The desired bit rateand associated quality of speech is highlyapplication dependent. Low bit rate: these basically have rates of between 75 and 2400 bps

(bitspersecond). Mediumtohigh bit rate: operate at greater than 2400 bps.

8/6/2019 T1 Intro Speech Processing

10/21

Tractament Digital de la Parla 10

Applications of Speech Coding

Reduction in bitrate for transmission/storage Speech enhancement (removal of noise)

Allows the development of applications for

Security

High definition TV

Teleconference

Etc.

8/6/2019 T1 Intro Speech Processing

11/21

SPEECH SYNTHESIS

8/6/2019 T1 Intro Speech Processing

12/21

Tractament Digital de la Parla 12

Linguistic analysis stage: mapsthe input text into a standard form;determines the structure of theinput, and finally decides how topronounce it.

Synthesis stage: converts thesymbolic representation of what tosay into an actual speechwaveform.

Speech Synthesis The aim of speech synthesis is to be able to take a word

sequence and produce human like speech

LinguisticAnalysis

SpeechSynthesis

Text

Prosody / Phone sequence

Sound Wave

8/6/2019 T1 Intro Speech Processing

13/21

Tractament Digital de la Parla 13

Applications of Speech Synthesis/Text-to-Speech (TTS)

Games Telephone-based Information

directions, air travel, banking, etc.

Accessing variable information

Machine-human interfaces Eyes-free (in car)

Reading/speaking for disabled

Reading of texts/books

Email access Education (Reading tutors)

Alarm systems

....

8/6/2019 T1 Intro Speech Processing

14/21

AUTOMATIC SPEECHRECOGNITION (ASR)

8/6/2019 T1 Intro Speech Processing

15/21

Tractament Digital de la Parla 15

Automatic Speech Recognition

Automatic Speech Recognition

(ASR) is the process ofconverting an unknown speechwaveform into thecorresponding orthographictranscription.

& language model

8/6/2019 T1 Intro Speech Processing

16/21

Extraction of feature vectors

Speech signalUsually every 10ms,25ms window

Acoustic featuresTypically around 39

8/6/2019 T1 Intro Speech Processing

17/21

Current issues in ASR

Steady reduction has been achieved over the last 20 yearsin many domains. Still more research is needed:

Increase robustness for new acoustic environments

Vocabulary increase and topic independence

Improve OOV (out-of-vocabulary) recognition

8/6/2019 T1 Intro Speech Processing

18/21

Tractament Digital de la Parla 18

Applications of SpeechRecognition/Understanding (ASR/ASU)

Dictation Telephone-based Information

directions, air travel, banking, etc

Polls, online shopping

Call routing

Hands-free

in car, computer, home(domotics), controlling tools

Second language (accent reduction)

Audio archive searching Help for disabled people

8/6/2019 T1 Intro Speech Processing

19/21

SPEAKER/LANGUAGERECOGNITION

8/6/2019 T1 Intro Speech Processing

20/21

Speaker/Language identification

Feature extraction

Speaker/language

models

Audio

Feature vectors

Selected speaker/language

8/6/2019 T1 Intro Speech Processing

21/21

Tractament Digital de la Parla 21

Applications of Speaker/LanguageRecognition

Language recognition for call routing

Speaker Recognition:

Speaker verification (binary decision) Voice password, telephone assistant

Speaker identification (one of N) (open set/closed set)

Criminal investigation