Georgi Tinchev

Georgi Tinchev, Marta Czarnowska, Kamil Deja et al. · amazon-science

This work focuses on modelling a speaker's accent that does not have a dedicated text-to-speech (TTS) frontend, including a grapheme-to-phoneme (G2P) module. Prior work on modelling accents assumes a phonetic transcription is available for the target accent, which might not be the case for low-resource, regional accents. In our work, we propose an approach whereby we first augment the target accent data to sound like the donor voice via voice conversion, then train a multi-speaker multi-accent TTS model on the combination of recordings and synthetic data, to generate the donor's voice speaking in the target accent. Throughout the procedure, we use a TTS frontend developed for the same language but a different accent. We show qualitative and quantitative analysis where the proposed strategy achieves state-of-the-art results compared to other generative models. Our work demonstrates that low resource accents can be modelled with relatively little data and without developing an accent-specific TTS frontend. Audio samples of our model converting to multiple accents are available on our web page.

Dariusz Piotrowski, Renard Korzeniowski, Alessio Falai et al.

In this work, we introduce a framework for cross-lingual speech synthesis, which involves an upstream Voice Conversion (VC) model and a downstream Text-To-Speech (TTS) model. The proposed framework consists of 4 stages. In the first two stages, we use a VC model to convert utterances in the target locale to the voice of the target speaker. In the third stage, the converted data is combined with the linguistic features and durations from recordings in the target language, which are then used to train a single-speaker acoustic model. Finally, the last stage entails the training of a locale-independent vocoder. Our evaluations show that the proposed paradigm outperforms state-of-the-art approaches which are based on training a large multilingual TTS model. In addition, our experiments demonstrate the robustness of our approach with different model architectures, languages, speakers and amounts of data. Moreover, our solution is especially beneficial in low-resource settings.

Yunlong Jiao, Adam Gabrys, Georgi Tinchev et al.

We present a universal neural vocoder based on Parallel WaveNet, with an additional conditioning network called Audio Encoder. Our universal vocoder offers real-time high-quality speech synthesis on a wide range of use cases. We tested it on 43 internal speakers of diverse age and gender, speaking 20 languages in 17 unique styles, of which 7 voices and 5 styles were not exposed during training. We show that the proposed universal vocoder significantly outperforms speaker-dependent vocoders overall. We also show that the proposed vocoder outperforms several existing neural vocoder architectures in terms of naturalness and universality. These findings are consistent when we further test on more than 300 open-source voices.

Georgi Tinchev, Shuda Li, Kai Han et al.

In this paper, we aim at establishing accurate dense correspondences between a pair of images with overlapping field of view under challenging illumination variation, viewpoint changes, and style differences. Through an extensive ablation study of the state-of-the-art correspondence networks, we surprisingly discovered that the widely adopted 4D correlation tensor and its related learning and processing modules could be de-parameterised and removed from training with merely a minor impact over the final matching accuracy. Disabling these computational expensive modules dramatically speeds up the training procedure and allows to use 4 times bigger batch size, which in turn compensates for the accuracy drop. Together with a multi-GPU inference stage, our method facilitates the systematic investigation of the relationship between matching accuracy and up-sampling resolution of the native testing images from 1280 to 4K. This leads to discovery of the existence of an optimal resolution $\mathbb{X}$ that produces accurate matching performance surpassing the state-of-the-art methods particularly over the lower error band on public benchmarks for the proposed network.

Milad Ramezani, Georgi Tinchev, Egor Iuganov et al.

In this paper, we present a factor-graph LiDAR-SLAM system which incorporates a state-of-the-art deeply learned feature-based loop closure detector to enable a legged robot to localize and map in industrial environments. These facilities can be badly lit and comprised of indistinct metallic structures, thus our system uses only LiDAR sensing and was developed to run on the quadruped robot's navigation PC. Point clouds are accumulated using an inertial-kinematic state estimator before being aligned using ICP registration. To close loops we use a loop proposal mechanism which matches individual segments between clouds. We trained a descriptor offline to match these segments. The efficiency of our method comes from carefully designing the network architecture to minimize the number of parameters such that this deep learning method can be deployed in real-time using only the CPU of a legged robot, a major contribution of this work. The set of odometry and loop closure factors are updated using pose graph optimization. Finally we present an efficient risk alignment prediction method which verifies the reliability of the registrations. Experimental results at an industrial facility demonstrated the robustness and flexibility of our system, including autonomous following paths derived from the SLAM map.

Georgi Tinchev, Adrian Penate-Sanchez, Maurice Fallon

We present SKD, a novel keypoint detector that uses saliency to determine the best candidates from a point cloud for tasks such as registration and reconstruction. The approach can be applied to any differentiable deep learning descriptor by using the gradients of that descriptor with respect to the 3D position of the input points as a measure of their saliency. The saliency is combined with the original descriptor and context information in a neural network, which is trained to learn robust keypoint candidates. The key intuition behind this approach is that keypoints are not extracted solely as a result of the geometry surrounding a point, but also take into account the descriptor's response. The approach was evaluated on two large LIDAR datasets - the Oxford RobotCar dataset and the KITTI dataset, where we obtain up to 50% improvement over the state-of-the-art in both matchability and repeatability. When performing sparse matching with the keypoints computed by our method we achieve a higher inlier ratio and faster convergence.

Georgi Tinchev, Adrian Penate-Sanchez, Maurice Fallon

Localization in challenging, natural environments such as forests or woodlands is an important capability for many applications from guiding a robot navigating along a forest trail to monitoring vegetation growth with handheld sensors. In this work we explore laser-based localization in both urban and natural environments, which is suitable for online applications. We propose a deep learning approach capable of learning meaningful descriptors directly from 3D point clouds by comparing triplets (anchor, positive and negative examples). The approach learns a feature space representation for a set of segmented point clouds that are matched between a current and previous observations. Our learning method is tailored towards loop closure detection resulting in a small model which can be deployed using only a CPU. The proposed learning method would allow the full pipeline to run on robots with limited computational payload such as drones, quadrupeds or UGVs.

Georgi Tinchev, Simona Nobili, Maurice Fallon

In this work we introduce Natural Segmentation and Matching (NSM), an algorithm for reliable localization, using laser, in both urban and natural environments. Current state-of-the-art global approaches do not generalize well to structure-poor vegetated areas such as forests or orchards. In these environments clutter and perceptual aliasing prevents repeatable extraction of distinctive landmarks between different test runs. In natural forests, tree trunks are not distinctive, foliage intertwines and there is a complete lack of planar structure. In this paper we propose a method for place recognition which uses a more involved feature extraction process which is better suited to this type of environment. First, a feature extraction module segments stable and reliable object-sized segments from a point cloud despite the presence of heavy clutter or tree foliage. Second, repeatable oriented key poses are extracted and matched with a reliable shape descriptor using a Random Forest to estimate the current sensor's position within the target map. We present qualitative and quantitative evaluation on three datasets from different environments - the KITTI benchmark, a parkland scene and a foliage-heavy forest. The experiments show how our approach can achieve place recognition in woodlands while also outperforming current state-of-the-art approaches in urban scenarios without specific tuning.