How do you sound Standards for Objective Speech Measurement

1
How do you sound?Standards for Objective Speech
Measurement
2
Agenda

• About objective speech measurement
• Speech measurement standards - understanding the
  differences between
• PSQM
• PSQM
• PAMS
• PESQ
• RADCOM support

3
Objective Speech Measurement

• Speech codecs are becoming common in many audio
  communications systems
• Modern communications networks cannot be reliably
  assessed by conventional engineering metrics such
  as signal-to-noise ratio
• Subjective tests (expensive, slow and unusable in
  applications like service monitoring)

4
Objective Speech Measurement (2)

• Objective models, based on human perception, were
  developed with the aim of predicting the results
  of subjective tests
• Complex algorithms were developed to emphasize
  only those parts of the input speech signal of
  interest to human listeners

5
Well-Known Objective Standards

• PSQM, PSQM Perceptual Speech Quality developed
  by KPN ITU-T P.861
• PAMS Perceptual Analysis Measurement System
  developed by British Telecom
• Both are intrusive listening speech quality
  assessment tools
• PSQM output scores range from 0-6.5 the lower
  the score, the better the voice quality
• PAMS output scores range from 1-5 the higher the
  score, the better the voice quality

6
PSQM Standard

• Defined by ITU-T Recommendation P.861
• Evaluates quality of voice signals in the same
  way as non-linear codecs encode and decode voice
  signals
• Codec assessment models such as PSQM can be
  unreliable when used in certain applications,
  especially systems that include linear filtering
  and/or delay variations
• An enhanced version, PSQM, correlates more to
  MOS scores in the presence of network degradation

7
PSQM Standard (2)

• Algorithm compares the signal received after the
  coder/decoder process with the original signal

8
PSQM Standard

• Enhanced version of PSQM which correlates more to
  MOS scores in the presence of network degradation
• Algorithm performs some post-processing on the
  PSQM value based on PSQM and signal power
• Accounts for different perceptions of volume and
  loud distortions
• As speech signals vary greatly in time, large
  false errors can occur e.g., a delay change of 20
  mSec can cause PSQM/PSQM to measure a drop in
  quality equivalent to approx. 1 MOS
• (Source Draft of AES, Sept. 2000)

9
PAMS Standard

• Designed to provide a repeatable, objective means
  for measuring perceived voice quality
• Provides a "Listening Quality" and "Listening
  Effort" scores
• Signals in the model are divided into sections
  known as utterances
• Delay packet-base transmissions such as IP
  telephony are identified
• Computes two quality ACR (absolute category
  rating method) measures
• Listening quality opinion scale 5-1
• Listening effort opinion scale 5-1

10
PAMS Standard (2.)

• Algorithm performs variable delay time alignment
  and transfer function equalization for reliable
  subjective assessment

PAMS Structure
Align Equalize
AuditoryTransform
Reference Signal
Prediction of perceived listing quality and
listing effort
Regression
Error Parameterization
Degraded Signal
Auditory Transform
11
PESQ Standard (ITU-T P.862)

• PESQ Perceptual Evaluation of Speech Quality
• Result of several years of work to develop an
  objective speech quality measure applicable to
  both speech codecs and end-to-end measurements
• Listening quality rating 1-5
• Combines the strongest parts of PAMS and PSQM
  algorithms
• Best improvement combines PAMS time alignment
  routine with PSQM perceptual model
• Designed to take variable delay into account,
  showing much higher correlation than PSQM
• (Source Draft of AES, Sept. 2000)

12
RADCOM Support

• VoIP Performer supports PAMS and PESQ
• MediaPro PAMS and PESQ scores, ability to
  playback VoIP, voice calls and provide jitter
  loss statistics
• QPro intrusive listening measurement of
  background noise, round-trip delay, insertion
  loss and echo
• QPro Supports E1/T1, ISDN BRI and analog
  interfaces