2. Definition and Coordination of Signal Processing Functions for telephone connections involving automotive speakerphones Scott Pennock Senior Hands-Free Standards Specialist QNX Software Systems

3. 2 All content copyright QNX Software Systems Outline Introduction List of Signal Processing Functions (SPFs) Detail for each SPF Problem trying to solve SPF solution Coordination between terminal and network Speech Processing Equipment (SPE) Conclusions

4. 3 All content copyright QNX Software Systems Introduction Telephone connections involving automotive speakerphones are among the most challenging types of connections Signal Processing Functions (SPFs), also known as speech enhancement functions, are essential for user acceptance SPFs reduce or eliminate: Fundamental problems caused by Far-field transducers Noisy environments Terminal design problems Network-related problems

5. 4 All content copyright QNX Software Systems List of SPFs SPFs include: Send or Ingress (entering network) Acoustic Echo Cancellation (AEC) Line Echo Cancellation (LEC) Microphone Array Processing (MAP) High-Frequency Encoding (HFE) Noise Reduction (NR) Equalization (EQ) Automatic Level Control (ALC; also referred to as AGC) Limiting (LM) Receive or Egress (exiting network) Band-Width Extension (BWE) EQ ALC Noise Compensation (NC; also referred to as ALE) LM

6. 5 All content copyright QNX Software Systems Acoustic Echo Cancellation (AEC) Acoustic echo Far end talker hears an echo of his/her own voice Caused by sound coming out of the loudspeaker being picked-up by the microphone and transmitted back to the talker NRBWEEQALCNCLM ALCEQNRHFEMAP Network AEC Far end SPF solution  Acoustic Echo Cancellation (AEC) can eliminate acoustic echo heard by far end  Allows both ends to simultaneously talk without noticeable speech or noise level fluctuations  This is known as “Type 1” full duplex performance AEC Loudspkr Mic To network From network SPE coordination  Only SPE closest to source of acoustic echo should be used Echo path delay likely known and stable Echo path more linear Can be optimized to acoustic interface

7. 6 All content copyright QNX Software Systems Microphone Array Processing (MAP) Variable speech levels Far end hears speech level differences Caused by different positions of users relative to microphone High noise level Far end hears high levels of background or electrical noise Caused by far distance of microphone and high vehicle noise levels NRBWEEQALCNCLM ALCEQNRHFEMix Network Far end SPF Solution  Microphone Array Processing (MAP) intelligently combines signals from multiple mics to optimize speech pick-up and reduce noise  Should be robust to mic type and spacing AEC MAP Mic 2 Mic 1 Out SPE coordination  No coordination required because specific to acoustic interface

8. 7 All content copyright QNX Software Systems High Frequency Encoding (HFE) Poor intelligibility of narrowband calls Far end listener has difficulty correctly recognizing speech sounds Caused by the narrowband telephone network filtering out speech frequencies above around 3400Hz NRBWEEQALCNCLM ALCEQNRHFEMAP Network Far end SPF solution  High Frequency Encoding (HFE) improves intelligibility of narrowband calls  High-frequency speech energy is encoded into lower frequencies that are transmitted by network AEC HFE Mic Out HFE SPE coordination  Only SPE closest to bandwidth restriction should be used

9. 8 All content copyright QNX Software Systems Noise Reduction (NR) High noise level Far end hears high levels of noise Caused by: Far distance of microphone Acoustic noise sources Airflow NRBWEEQALCNCLM ALCEQNRHFEMAP Network Far end SPF solution  Noise Reduction (NR) reduces noise heard by the far end  Can be used to compensate for design problems  Important to track noise even during speech AEC NR Mechanical vibration Electro-Magnetic Interference (EMI) SPE coordination  Only SPE closest to signal source should be used

10. 9 All content copyright QNX Software Systems Equalization (EQ) Unnatural speech Far end listener hears speech that does not have the same timbre as the talker Caused by filtering of microphone and other components of the speech signal path NRBWEEQALCNCLM ALC WBNRHFEMAP Network Far end SPF Solution  Equalization (EQ) makes speech sound more like the person talking  Should use different EQ curves for low and high noise environments AEC EQ Mic Out EQ SPE coordination  SPE closest to acoustic interface should be used  EQ should be done before ALC

11. 10 All content copyright QNX Software Systems Automatic Level Control (ALC) Variable speech levels Far end hears speech level differences Caused by different speaking levels and distances of users relative to microphone NRBWEEQALCNCLM EQNRHFEMAP Network Far end SPF solution  ALC ensures far end hears a consistent perceived loudness by adjusting level  Should be robust to different signal types and only train on speech AEC ALC ALC “Off” ALC “On” SPE coordination  SPE closest to acoustic interface should be used  ALC should be done after other SPFs that will affect the speech signal level

12. 11 All content copyright QNX Software Systems Limiter (LM) Speech distortion Far end hears distorted speech (has “buzzy” quality) Caused by amplitude clipping in signal path (e.g., exceeding dynamic range of telephone system) NRBWEEQALCNCLM ALCEQNRHFEMAP Network Far end SPF solution  Limiter (Lim) reduces distorted speech caused by amplitude clipping  Protects against speech distortion caused from large peaks in speech AEC LM SPE coordination  SPE closest to acoustic interface should be used

13. 12 All content copyright QNX Software Systems Noise Reduction (NR) on Receive Noise Near end listener hears noise Caused by some Network Access Device (NAD) chips May only be audible in speakerphone mode because of increased gain NRBWEEQALCNCLM ALCEQNRHFEMAP Network Far end SPF solution  NR eliminates objectionable noise heard by near end AEC NR SPE coordination  Should only be run when noise detected  Needs to be run in tandem with NR in SPE at far end acoustic interface  Need NR at far end of connection to help with speech encoding  Need NR at near end of connection because receive noise generated after far end NR

14. 13 All content copyright QNX Software Systems BandWidth Extension (BWE) Narrowband call quality Near end listener complains of poor quality of narrowband calls Caused by contrast effect with higher quality wideband calls and increasing user expectations due to exposure to wideband quality NRBWEEQALCNCLM ALCEQNRHFEMAP Network Far end SPF solution  BWE improves quality of narrowband calls and helps minimize contrast effect and user dissatisfaction  BWE reconstructs low and high frequency information removed by the telephone network based on speech information that was transmitted AEC BWE To loudspkr From network BWE SPE coordination  SPE closest to acoustic interface should be used

15. 14 All content copyright QNX Software Systems Equalization (EQ) Unnatural speech Near end listener hears speech that does not have the same timbre as the talker Caused by filtering of loudspeaker and other components of the speech signal path NRBWEEQALCNCLM ALCEQNRHFEMAP Network Far end SPF solution  EQ makes far end talker sound more like themself  Compensates for frequency response of loudspeaker in housing  Prevents loudspeaker distortion by attenuating frequencies which overdrive the loudspeaker AEC EQ SPE coordination  SPE closest to acoustic interface should be used

16. 15 All content copyright QNX Software Systems Automatic Level Control (ALC) Variable speech levels Near end hears speech level differences across calls There are several causes: Bandwidth differences between narrowband and wideband calls Speaking level differences across talkers at far end Other fixed gains that vary from call to call (e.g., far end terminal, network, Bluetooth, etc.) NRBWEEQAGCNCLM ALCEQNRHFEMAP Network Far end SPF solution  ALC ensures near end hears a consistent perceived loudness  Should account for effects of bandwidth (wideband vs. narrowband) and the dynamic nature of speech AEC ALC SPE coordination  SPE closest to acoustic interface should be used

17. 16 All content copyright QNX Software Systems Noise Compensation (NC) Variable speech levels Near end hears speech level differences across noise conditions Caused by a perceptual effect known as loudness masking in human auditory system NRBWEEQALCDLCLM ALCEQNRHFEMAP Network Far end SPF solution  Noise Compensation (NC) causes near end to hear speech at the same loudness across noise conditions  Automatically adjusts receive volume control based on noise measurement—which is better than wheel-speed  Improves intelligibility in the presence of noise AEC NC NC “Off” NC “On” SPE coordination  SPE closest to acoustic interface should be used

18. 17 All content copyright QNX Software Systems Limiter (LM) Speech distortion Near end hears distorted speech (has “buzzy” quality) Caused by overdriving digital domain, electronic components, or loudspeaker Acoustic echo or speech level fluctuations Far end hears echo of his/her own voice and/or speech level fluctuations of near end talker Caused by overdriving system components which leads to echo path distortion and ultimately poor AEC performance NRBWEEQALCNCLim LMALCEQNRHFEMAP Network Far end SPF solution  The Limiter reduces or eliminates speech distortion, echo, and speech level fluctuations caused by overdriving system components  It does this by soft limiting the receive speech signal before it is hard-clipped AEC LM SPE coordination  SPE closest to acoustic interface should be used

19. 18 All content copyright QNX Software Systems Line Echo Canceller (LEC) ITU-T G.799.2/Figure 2  Block diagram of an echo canceller [ITU-T G.168] Network echo Near end talker hears an echo of his/her own voice Caused by electrical reflection of speech energy at network hybrid SPF solution  Line Echo Cancellation (LEC) can eliminate acoustic echo heard by Near end  Allows both ends to simultaneously talk without noticeable speech or noise level fluctuations SPE coordination  Only SPE closest to source of network echo should be used

20. 19 All content copyright QNX Software Systems Conclusions SPFs are essential for user acceptance of telephone connections involving automotive speakerphones Where SPFs are placed along the telephone connection is important to their effectiveness Running SPFs in tandem can degrade performance It is important to coordinate the operation of SPFs along the connection

21. 20 All content copyright QNX Software Systems Thank you Scott Pennock Senior Hands-Free Standards Specialist spennock@qnx.com