This work introduces a method for speech intelligibility testing in reproduced sound scenes. The proposed method uses background sound scenes augmented by target speech sources and reproduced over a multichannel loudspeaker setup with time-frequency domain parametric spatial audio techniques. Subjective listening tests were performed to validate the proposed method: speech recognition thresholds (SRT) in noise were measured in a reference sound scene and in a room where the reference was reproduced by a loudspeaker setup. The listening tests showed that for normally-hearing test subjects the method provides nearly indifferent speech intelligibility compared to the real-life reference when using a nine-loudspeaker reproduction setup in anechoic conditions (<0.3 dB error in SRT). Due to the flexible technical requirements, the method is potentially applicable to clinical environments.
展开▼