2 nd Workshop on Wideband Speech Quality in
2 nd Workshop on Wideband Speech Quality in Terminals and Networks: Assessment and Prediction 22 nd and 23 rd June 2005 - Mainz, Germany A new codec within ITU-T: new bandwidth Catherine Quinquis - France Telecom R&D Vice-chair of 3 GPP SA 4 and co-rapporteur of ITU-T Q 7/12 catherine. quinquis@francetelecom. com 2 nd Workshop on Wideband Speech Quality - June 2005 1
A new codec within ITU-T: G 722. 1 Annex C q Extension toward higher frequency of existing transform codec G. 722. 1 Ø Input/output signal bandwidth : [50 Hz; 14 k. Hz] q The foreseen applications for the extension codec are: Ø Video conferencing Ø Hands-free teleconferencing Ø And internet streaming audio q Qualification in November 2004 q Only one candidate (that passes the requirements) Characterisation phase q Approval date : April 2005 2 nd Workshop on Wideband Speech Quality - June 2005 2
A new codec within ITU-T: G 722. 1 Annex C Characterisation methodology q Phase 1 dealing with the main foreseen application: audio conferencing Ø Clean speech (ACR method) • Case of high quality sound recording with minimised room effect Ø Reverberant speech with background noise (DCR method) • Case of typical audio conferencing systems (different room sizes, distances to microphone and types of background noise) q Phase 2 dealing with optional application : audio streaming Ø Mushra methodology • Selected Anchors: 7 k. Hz and 10 k. Hz band limited signals Ø Music • Classical, jazz, modern, singer, orchestral Ø Mixed content items • Advertisement, film trailer, news music announcement 2 nd Workshop on Wideband Speech Quality - June 2005 3
A new codec within ITU-T: G 722. 1 Annex C Requirements q Terms of reference Ø Bit rates : 24 kb/s, 32 kb/s and 48 kb/s q Quality requirements Ø Clean speech • Not worse than MPEG 4 AAC-LD for the same bit rate at 99% confidence interval Ø Reverberant speech with background noise • Not worse than MPEG 4 AAC-LD for the same bit rate at 99% confidence interval q Characterisation Ø Comparison at 95 % confidence interval to avoid problems with Type I and type II errors. • Type I error: reject a true null hypothesis (5% for 95% CI) • Type II error: fail to reject a false null hypothesis • The smallest Type I error is, the larger type II errors will be Ø Add comparison to existing 3 GPP audio codecs operating at same bit rate • Audio codec for MMS, PSS and MBMS services – Extended AMRWB (3 GPP TS 26. 304) – Enhanced aac. Plus (3 GPP TS 26. 410) • Comparison limited to 24 and 32 kbps, 3 gpp audio codecs did operate above 36 kbps in mono. 2 nd Workshop on Wideband Speech Quality - June 2005 4
A new codec within ITU-T: G 722. 1 Annex C Phase 1 Characterisation results (1/4) q Clean Speech Ø All requirements are met; even better than MPEG 4 AACLD at low bit rates (24 and 32 kbps) 2 nd Workshop on Wideband Speech Quality - June 2005 5
A new codec within ITU-T: G 722. 1 Annex C Phase 1 Characterisation results (2/4) q Reverberant speech with office noise Ø Reverberation: Small room (5 people) and microphone at 50 cm Ø All requirements are met; even better than MPEG 4 AACLD at low bit rates(24 and 32 kbps) 2 nd Workshop on Wideband Speech Quality - June 2005 6
A new codec within ITU-T: G 722. 1 Annex C Phase 1 Characterisation results (3/4) q Reverberant speech with interfering talker Ø Reverberation: medium room (25 people) and microphone at 50 cm Ø All requirements are met; even better than MPEG 4 AACLD at low bit rates(24 and 32 kbps) 2 nd Workshop on Wideband Speech Quality - June 2005 7
A new codec within ITU-T: G 722. 1 Annex C Phase 1 Characterisation results (4/4) q Reverberant speech with office noise and interfering talker Ø Reverberation: medium room (25 people) and microphone at 100 cm Ø All requirements are met; even better than MPEG 4 AACLD at low bit rates(24 and 32 kbps) 2 nd Workshop on Wideband Speech Quality - June 2005 8
A new codec within ITU-T: G 722. 1 Annex C Phase 2 Characterisation results (1/3) q 24 kbps Ø better than MPEG 4 AAC-LD and about the same as 3 GPP codecs Ø G 722. 1 annex C and MPEG 4 AAC-LD and Enhanced aac. Plus are more sensitive to content type than extended AMRWB 2 nd Workshop on Wideband Speech Quality - June 2005 9
A new codec within ITU-T: G 722. 1 Annex C Phase 2 Characterisation results (2/3) q 32 kbps Ø better than MPEG 4 AAC-LD and about the same as 3 GPP codecs Ø G 722. 1 annex C and MPEG 4 AAC-LD and Enhanced aac. Plus are more sensitive to content type than extended AMRWB 2 nd Workshop on Wideband Speech Quality - June 2005 10
A new codec within ITU-T: G 722. 1 Annex C Phase 2 Characterisation results (3/3) q 48 kbps Ø better than MPEG 4 AAC-LD Ø 3 GPP codecs do not have 48 kbps with mono input signal. 2 nd Workshop on Wideband Speech Quality - June 2005 11
A new codec within ITU-T: G 722. 1 Annex C Extension of P. 341 q P. 341 Ø Need to extend the P. 341 recommendation to the super wide band Ø filter simulating the terminal response has been made available in the STL with ITU-T SG 16; this simulation was used for the characterization testing exercise. P 341 Extension: upper and lower limits Frequency (Hz) P 341 and its Extension Frequency (Hz) 2 nd Workshop on Wideband Speech Quality - June 2005 12
A new codec within ITU-T: G 722. 1 Annex C E model q E model Ø An extension to wide band (50 Hz- 7 k. Hz) is been discussed Ø A new bandwidth (50 Hz – 14 k. Hz) is going to be used in the network: it should also be taken into account. Ø Only one codec has been tested in this bandwidth with some adaptation on MNRU range 2 nd Workshop on Wideband Speech Quality - June 2005 13
A new codec within ITU-T: G 722. 1 Annex C MNRU curve q MNRU adaptation Ø to get more score outside the saturation region, in G 722. 1 C characterisation test, lowest MNRU is 18 d. B 2 nd Workshop on Wideband Speech Quality - June 2005 14
A new codec within ITU-T: G 722. 1 Annex C Thank you 2 nd Workshop on Wideband Speech Quality - June 2005 15
- Slides: 15