Applied Phonetics - Quiz 2

L2 Accents

aspects of speech that identify someone as a second language speaker

L2 accents = highly salient

its is often easy to identify second language speakers from speech patterns alone

Segmentals

individual consonants, vowels

Suprasegmentals

word stress, prominence, rhythm, intonation

Affrication

occurs when a consonant changes its manner of articulation to become an affricate

Perceptual Dimension of L2 Speech evaluation: Accentedness

Perceived difference from local variety

Perceptual Dimension of L2 speech Evaluation: Comprehensibility

the listener's effort

Perceptual Dimension of L2 Speech Evaluation: Intelligibility

how much is understood

Perceptual Dimension of L2 speech evaluation: Fluency

speech and the flow of speech

Key factors that affect learners’ pronunciation

• L1 influence

• age

• language proficiency

• exposure to the target language

• length of residence

• attitudes towards pronunciation

• strength of in-group identification

L2 Pronunciation/Accent Modification: Intelligibility principle

“Different features have different effects on understanding. Instruction should focus on those features that are most helpful for understanding and should deemphasize those that are relatively unhelpful”.

L2 Pronunciation/Accent Modification: Nativeness Principle

“it is both possible and desirable to achieve native-like pronunciation in a foreign language

What causes a loss of intelligibility?

global aspects of speech production (suprasegmentals, voice) affect intelligibly

generalizability

how well students can perceive/produce phonemes in different phonetic contexts or when produced by different talkers.

Speech Production Model: conceptualization

disfluencies occur at sentences/clause boundaries

Speech Production Model: Formulation

differences occur within sentences/clauses

Speech Production Model: Articulation

disfluencies occur or within individual words/short phrases

Speech Production Model: Monitoring

disfluencies may be compounded by self-monitoring, causing disfluencies to happen at any stage.

Classroom tasks to help enhance fluency

• awareness-raising activities - training students to give each other focused feedback

• formulaic language instruction - the majority of speech is formulaic in nature.

Speech Synthesis

the process of creating utterances that are generated entirely or partially without a human vocal tract

Strategies to create synthetic utterances

• articulatory - based on the structure of the vocal tract (power source, source filter)

• Acoustic - based on the acoustic characteristics (e.g., duration, pitch, formants)

• Concatenative - real speech segments are combined in sequence.

history of speech synthesis

1. mechanical devices

2. analog electrical/electronic devices

3. digital (software-based) synthesizers

Characteristics of “good” synthetic speech

• intelligible

• easy to process

• “stand up” to noise

• natural (human-like)

Dimensions of Speech

• intelligibility - how much you understand

• comprehensibility - how easy it is to process

• naturalness - synthetic speech that sounds natural

what happens during the intelligibility assessment

• present synthetic utterances (word, phrases, sentences) to listeners

• have them write down what they hear

• % correct words average over the listeners = the intelligibility score of the speech

activities for forensic phoneticians

• earwitness speaker identification

• expert speaker identification

• speaker profiling

when to use earwitness identification

when the face of the perpetrator of a crime is not seen; however the voice is heard by an earwitness

• investigators and prosecutors must rely on the eyewitnesses ability to recall the perpetrators voice

problems with creating an earwitness line up

• foil voices (not suspects) must be similar enough that the suspect does not stand out

• voices should have equal probability of being selected by a non-witness.

speaker recognition

the speaker of an utterance is determined

acoustic methods

forensic phoneticians and slp’s learn to “read” spectrograms and waveforms with a degree of fluency.

what makes FSID difficult?

• poor quality (noisy) recordings

• effects of emotions on speech

• speech disguise

speaker profiling

attempting to determine characteristics of the perpetrator from speech patterns

• involves analysing the sound of speech to determine something about the speaker

automatic speech recognition (ASR)

an independent, machine-based process of decoding and transcribing oral speech

speech recognition

the process of determining the meaning of an utterance

speaker recognition

the speaker of an utterance is determined (might be done by a computer)

characteristics of ASR systems

• signal processing

• natural language processing

• machine learning

three dimensions of ASR Systems: speaker dependence

• speaker-dependent - trained for each individual speaker

• speaker-independent - training database is large enough to account for new speakers

three dimensions of ASR systems: speech continuity

• discrete word recognition systems - recognizes individual words in isolation

• connected word recognition systems - recognizes individual words with pauses between

• continuous speech recognition systems - recognizes whole sentences without delibrate pauses between word

three dimensions of ASR systems: Vocabulary size

• small/large

Potential Errors with ASR systems

• deletion - system does not receive the word

• addition - system perceives noise as speech unit

• substitution - system perceives a non-existent phonemic substitution

• rejection - system rejects the word because it is not part of its vocabulary list

• one speech unit combines into two

• false alarms - word is misidentified