In Mandarin Chinese, the fundamental frequency (F0) contour defines lexical “Tones” that differ in meaning despite being phonetically identical. Flattening the F0 contour impairs the intelligibility of Mandarin Chinese in background sounds. This might occur because the flattening introduces misleading lexical information. To avoid this effect, two types of speech were used: single-Tone speech contained Tones 1 and 0 only, which have a flat F0 contour; multi-Tone speech contained all Tones and had a varying F0 contour. The intelligibility of speech in steady noise was slightly better for single-Tone speech than for multi-Tone speech. The intelligibility of speech in a two-talker masker, with the difference in mean F0 between the target and masker matched across conditions, was worse for the multi-Tone target in the multi-Tone masker than for any other combination of target and masker, probably because informational masking was maximal for this combination. The introduction of a perceived spatial separation between the target and masker, via the precedence effect, led to better performance for all target-masker combinations, especially the multi-Tone target in the multi-Tone masker. In summary, a flat F0 contour does not reduce the intelligibility of Mandarin Chinese when the introduction of misleading lexical cues is avoided.

This study examines whether speech rhythm affects speech recognition under “cocktail-party” conditions. Against a two-talker masker, but not a speech-spectrum noise masker, recognition of the last (third) keyword in a normal rhythmic sentence was significantly better than that of the first keyword. However, this word-position-related speech-recognition improvement disappeared for rhythmically hybrid target sentences that were constructed by grouping parts from different sentences with different artificially modulated rhythms (rates) (fast, normal, or slow). Thus, the normal rhythm with a constant rate plays a role in improving speech recognition against informational speech masking, probably through a build-up of temporal prediction for target words.

The performance of higherorder Ambisonics (HOA) signals obtained using spherical harmonics decomposition method is disturbed by two primary sources of errors, the noise pollution in low-frequency band and the spatial aliasing in high-frequency band. Inspired by the HOA signals upscale method, which is performed using the sparse character of the sound field, this paper propose a sound field decomposition model based on a sparse deep neural network that offers HOA signals with wider frequency bandwidth. We use the frequency domain multi-scale convolutional network to realize the spherical harmonics decomposition, as well as learning the spatial aliasing pattern, based on which the aliasing-free HOA signals can be derived. Besides, we apply a sparse encoding network to cpature the sparse feature of the sound field which will improve the model performance when the sparse condition is satisfied. The experiments results prove that the proposed model can obtain HOA signals with wider frequency range of operation under multiple sources (up to 10 sources) and low reverberant environments ( $T_{60}\le$ 400 ms). When the sparsity feature cannot be satisfied ( $T_{60} =$ 800 ms), the proposed network model still maintain the same performance as the traditional methods.