Introduction: In clinical practice, speech understanding in noise is most often evaluated using co-located speech and noise (i.e., without spatial cues). The French Matrix test was indeed validated in this condition of administration. Yet, most hearing aid centers in France are equipped with fiveloudspeaker arrays, which allows for speech to be presented from a single loudspeaker directly in front of the listener (0° azimuth) and noise to simultaneously presented from 4 loudspeakers (45°, 135°, 225°, and 315° azimuth; "diffuse noise"). This configuration is meant to be more representative of «ecological condition». Material and methods: We designed a prospective observational study in normal hearing (NH) and hearing impaired (HI) adults. Data were collected between June 2019 and December 2020 and consisted in: pure-tone air conduction thresholds for each ear using headphones for audiometric frequencies 0.5, 1, 2, 4, 6, and 8 kHz. Unaided monosyllable word recognition in quiet was measured in sound field at 65 dBA using monosyllable words from Lafon, in binaural condition. Unaided sentence recognition in noise was measured in sound field using the French Matrix (binaural performance). Results: 297 listeners across 9 hearing centers completed the study. Unaided SRTs in diffuse noise were significantly correlated with PTA thresholds, age at testing, as well as word and phoneme recognition scores in quiet. Stepwise linear regression analysis showed that SRTs in diffuse noise were significantly predicted by a combination of PTA threshold and word recognition scores in quiet. SRTs were measured in co-located and diffuse noise for 65 listeners (13 NH, 22 Mild HL, 19 Mod HL1, and 11 Mod HL2). With co-located noise, the mean SRT and standard deviation was 4.2±1.1, 0.2±3.2, 2.9±2.6, and 8.4±7.0 dB for the NH, Mild HL, Mod HL1, and Mod HL2 groups, respectively. With diffuse noise, the mean SRT and standard deviation was 8.9±1.4, 2.4±4.6, 1.9±4.3, and 8.8±9.2 dB for the NH, Mild HL, Mod HL1, and Mod HL2 groups, respectively. A RM ANOVA with noise type (diffuse, co-located) as the within-subject factor and hearing status group (NH, Mild HL, Mod HL1, Mod HL2) as the between-subject factor showed significant effects for noise type [F(1, 61)=30.3, p<0.001] and subject group [F(3, 61)=15.3, p<0.001]; there was a signifi- cant interaction [F(3, 46)=4.1, p=0.010]. Post-hoc Bonferroni pairwise comparisons showed that SRTs in diffuse noise were significantly lower than SRTs in co-located noise only for the NH and Mild HL groups (p<0.05 in both cases). For diffuse noise, there were significant differences in SRTs among all groups; for co-located noise, there was a significant difference in SRTs among all groups, except for between the Mild HL and Mod HL1 groups. Linear regression analysis showed a significant correlation between SRTs in co-located and diffuse noise (r2=0.71, p<0.001). Conclusions: The results are consistent with previous studies that found that hearing-impaired listeners had greater difficulty using spatial cues to segregate speech and noise. While diffuse noise may not be suitable for clinical evaluation of speech understanding in noise in hearing- impaired listeners, it may provide additional insights into their "real-world" speech understanding in complex listening environments. Moreover, as our study provides the range of possible values of unaided SRTs, they may serve as references for comparison with aided SRTs, more likely to be measured in a diffuse noise setup than in a co-located noise one. [ABSTRACT FROM AUTHOR]