Dolby Laboratories

About: Dolby Laboratories is a company organization based out in Amsterdam, Netherlands. It is known for research contribution in the topics: Audio signal & Audio signal flow. The organization has 956 authors who have published 1726 publications receiving 29456 citations.

SESQA: Semi-Supervised Learning for Speech Quality Assessment

Abstract: Automatic speech quality assessment is an important, transversal task whose progress is hampered by the scarcity of human annotations, poor generalization to unseen recording conditions, and a lack of flexibility of existing approaches. In this work, we tackle these problems with a semi-supervised learning approach, combining available annotations with programmatically generated data, and using 3 different optimization criteria together with 5 complementary auxiliary tasks. Our results show that such a semi-supervised approach can cut the error of existing methods by more than 36%, while providing additional benefits in terms of reusable features or auxiliary outputs. Improvement is further corroborated with an out-of-sample test showing promising generalization capabilities.

System and Method of Speaker Cluster Design and Rendering

Abstract: A method of outputting audio in a teleconferencing environment includes receiving audio streams, processing the audio streams according to information regarding effective spatial positions, and outputting, by at least three speakers arranged in more than one dimension, the audio streams having been processed. The information regarding the plurality of effective spatial positions corresponds to a perceived spatial scene that extends beyond the speakers in at least two dimensions. In this manner, participants in the teleconference perceive the audio from the remote participants as originating at different positions in the teleconference room.

Apparatus for scanning variable area optical sound tracks by parallel processing discrete photosensor outputs

Abstract: Apparatus for scanning variable area optical sound tracks wherein a plurality of discrete photo sensors examine small fractional portions of the sound track width. Selection of the number of sensors in relation to photographic grain dimensions results in converting annoying quantizing noise to more tolerable random noise while using a practical number of sensors. Electronic processing of the sensors' outputs reduces impulse noise from dirt and scratches.

Sums of Five Almost Equal Prime Squares

Abstract: Let P i , 1 ≤ i ≤ 5, be prime numbers. It is proved that every integer N that satisfies N ≡ 5(mod24) can be written as $$ N = p^{2}_{1} + p^{2}_{2} + p^{2}_{3} + p^{2}_{4} + p^{2}_{5} ,{\text{where}}{\left| {{\sqrt N }5 - p_{i} } \right|} \leqslant N^{{\frac{1} {2} - \frac{{19}} {{850}} + \in }} $$ .

Guided post-prediction filtering in layered VDR coding

Abstract: A visual dynamic range (VDR) coding system creates a sequence of VDR prediction images using corresponding standard dynamic range (SDR) images and a prediction function. For each prediction image, an encoder identifies one or more areas within the prediction image suitable for post-prediction filtering. For each identified post-prediction area, a post-prediction filtering mode is selected among one or more post-prediction filtering modes. The selected post-prediction filtering mode is applied to output a filtered prediction image. Information related to the post-prediction filtering areas and the selected corresponding post-prediction filtering modes may be communicated to a receiver (e.g., as metadata) for guided post-prediction filtering. Example post-prediction filtering modes that use low-pass averaging filtering or adaptive linear interpolation are also described.