The interpretation of genomic sequences is crucial for understanding biological processes. To handle the growing volume of DNA sequence data, Genomic Foundation Models (GFMs) have been developed by adapting architectures and training paradigms from Large Language Models (LLMs). Despite their remarkable performance in DNA sequence classification tasks, there remains a lack of systematic understanding regarding the training and task-adaptation processes of GFMs. Moreover, existing GFMs cannot achieve state-of-the-art performance on both short and long-context tasks and lacks multimodal abilities. By revisiting pre-training architectures and post-training techniques, we propose **Omni-DNA**, a family of models spanning 20M to 1.1B parameters that supports sequence understanding, long-context genomic reasoning, and natural-language annotation. **Omni-DNA** establishes new state-of-the-art results on 18 of 26 evaluations drawn from Nucleotide Transformer and Genomic Benchmarks. When jointly fine-tuning on biologically related tasks, **Omni-DNA** consistently outperform existing models and demonstrate multi-tasking abilities. To enable processing of arbitrary sequence lengths, we introduce **SEQPACK**—an adaptive compression operator that packs historical tokens into a learned synopsis using a position-aware learnable sampling mechanism, enabling transformer-based models to process ultra-long sequences with minimal memory and computational requirements. Our approach demonstrates superior performance on enhancer-target interaction tasks, capturing distant regulatory interactions at the 450kbp range more effectively than existing models. Finally, we present a new dataset termed **seq2func**, enabling Omni-DNA to generate accurate and functionally meaningful interpretations of DNA sequences, unlocking new possibilities for genomic analysis and discovery.

This study investigated the effects of loudness and smiling on timbre of speech in different languages and genders. A group of native speakers of Mandarin, German, and Danish (10 participants per language) were recorded while blocks of 10 sentences in 2x3 block-wise randomized loudness and smiling conditions. Four-way ANOVAs were conducted on 10 timbre parameters, including F1, F2, F3, H1-H2, H1-A3, HNR, CPP, CoG, jitter, and shimmer. Results showed that loudness affected more timbre features than smiling. Mandarin differed from German and Danish in such timbre features such as HNR and CPP, modulated by loudness and smiling. There were no acoustics interactions between loudness and smiling in any of the three languages.

This work investigated crosslinguistic perception of Mandarin utterances conveying six classes of attitudes, i.e., dominant/ submissive, friendly/hostile, polite/rude, serious/joking, praising/blaming, and sincere/insincere. Five groups of subjects were tested: native Mandarin speakers, Japanese L2 learners of Mandarin, French L2 learners of Mandarin, naive Japanese without Mandarin ability, and naive French without Mandarin ability. A set of Mandarin attitudinal utterances elicited in role-play dialogues were used as stimuli. Perceptual experiments showed that native subjects performed the best in identifying attitudes, and L2 learners judged better than naive foreigners. A complex interaction was found between attitude and the listener’s L1/L2 experience which was also closely related to culture. Also, the correlations between perceptual patterns and prosodic features were examined, and the results suggested that the prosodic cues for certain attitudes might be dependent on the listener’s language/culture background.

Scatterplots are widely used to visualize scatter dataset for exploring outliers, clusters, local trends, and correlations. Depicting multi-class scattered points within a single scatterplot view, however, may suffer from heavy overdraw, making it inefficient for data analysis. This paper presents a new visual abstraction scheme that employs a hierarchical multi-class sampling technique to show a feature-preserving simplification. To enhance the density contrast, the colors of multiple classes are optimized by taking the multi-class point distributions into account. We design a visual exploration system that supports visual inspection and quantitative analysis from different perspectives. We have applied our system to several challenging datasets, and the results demonstrate the efficiency of our approach.

We used a command-response additive F0 model to analyze F0 patterns of Japanese spoken by native speakers of Mandarin Chinese. Compared to native speakers of Japanese, we found that Chinese speakers exhibit the following characteristics: (a) higher pitch, (b) more phrases, (c) bunsetsu decomposition, and (d) utterance-final plunging. These characteristics physically manifest themselves as: (a) higher baseline F0, (b) more phrase commands, (c) more accent commands, and (d) negative commands. These characteristics may be subjectively perceived as: (a) tinnier speech (possible L1 marker but does not degrade communication), (b) disjoint phrases (requires mental consolidation), (c) choppy prosodic words (requires reconstruction), and (d) abrupt utterance termination (possibly misconstrued as emphatic or rude). We believe these difficulties arose from tonal and syllable-timed interference, which can be overcome by prosodic control and planning. Index Terms: F0 contour, command-response model, L2 learning, Japanese, accent, phrase.

Emphasis and question are two factors that have significant effects on F/sub 0/ contours for various languages, among which tone languages require more careful study because their F/sub 0/ contours show complex interaction between lexical tones and sentence intonation. This paper employs the command-response model for the process of F/sub 0/ contour generation to investigate the effects of these two factors for Cantonese, a typical tone language with nine tones. Analysis shows that the major effect of emphasis is on phrase commands, whereas the polarity and the amplitude of the tone commands in the emphasized part are hardly affected so that the inherent tone command patterns are maintained. In the intonation question, the inherent tone command in the later part of the sentence-final syllable is always substituted by a positive tone command. The particle question, on the other hand, maintains the inherent tone command for the question particle. In both types of questions, a sentence-final phrase command is added or enhanced, and a particular ending tone command is attached, the amplitude of which can indicate the degree of inquisitive intention. By comparison, the effect of emphasis starts from the target part for emphasis but is not confined to it, whereas the effect of question is localized in the sentence-final part and especially concentrated within the ending syllable. Nevertheless, both of them can be represented in the framework of the command-response model, by which F/sub 0/ contours for expressive speech can be generated efficiently.

Cantonese is a well-known Chinese dialect with a quite complex tone system. We have successfully applied the command-response model to represent F/sub 0/ contours of Cantonese speech by defining a set of appropriate tone command patterns. It provides an efficient means to describe Cantonese F/sub 0/ contours with high accuracy. Both qualitative and quantitative descriptions of the command patterns of nine tones are presented. The various cues for identifying each tone are investigated, based on which a set of rules is derived to synthesize F/sub 0/ contours of Cantonese. Perceptual experiments have been conducted to test the validity of the rules for each tone and to evaluate the naturalness of synthetic speech based on those rules.

Word emphasis and question intonation have significant effects on F0 contours for various languages. This paper employs the command-response F0 model to investigate the effects of the two factors for Cantonese, a typical tone language with nine tones. Analysis shows that the main effect of word emphasis associated with narrow focus is not on tone command but on phrase command. An additional or a larger phrase command tends to occur immediately preceding the emphasized word. For the tone command of the emphasized word, duration may be lengthened but polarity and amplitude are hardly affected so that the inherent tone patterns are kept. Also, echo question will replace the tone command in the later part of the sentence-final syllable by a positive tone command of much larger amplitude than usual. Both the effects of the two factors can be embedded in the framework of our model.

A new method for generating sentence F 0 contours of Mandarin speech is proposed.The method assumes the F 0 contour generation process model, but generates the tone and phrase components in different ways and sums them to produce a sentence F 0 contour.The tone component is generated concatenating F 0 patterns of tone nuclei, which are predicted by a corpus-based scheme (binary decision trees).Experiments of F 0 contour generation were conducted by using 100 news utterances by a female speaker.The results showed that the method could generate F0 contours close to those of target speech.A perceptual evaluation was also conducted on the synthetic speech using the F0 contours generated by the method.An average score of 4.5 in a 5-point scale indicates the high naturalness, verifying the validity of the method.

As a major Chinese dialect, Cantonese is well known for its complex tone system. This paper applies the command-response model to represent the F0 contours of Cantonese speech. Analysis-by-Synthesis is conducted on both utterances of carrier sentences and utterances with less constrained structures, from which a set of appropriate tone command patterns is derived. By intrinsically incorporating the effects of tone coarticulation, word accentuation and phrase intonation, the model provides a high accuracy of approximation to the F0 contours of Cantonese, and hence serves as a much better method to quantitatively describe the continuous F0 contours than the traditional tone letter scale notation system. The constraints in timing and amplitude of tone commands are also investigated, which can be used for synthesis of F0 contours. 1.

In this paper we present a novel language-independent probabilistic model for automatic grapheme-to-phoneme and phoneme-to-grapheme conversion of words. In a fully unsupervised training procedure, two processes are applied; the transformation rules, which usually fail to provide the correct symbols, are eliminated, and new variable-length string transformation rules are defined improving the string transformation accuracy in the training data. In an iterative process the probabilistic transformation rules are updated in the direction of reducing the error rate of the transformed symbols. Long-term dependencies are defined automatically. Training and testing of the model was carried out on lexicon and natural language corpora of six European Languages. Accurate generalisations have been achieved in all experiments for both transformation directions using a relative small number of defined rules in the training procedure. It is demonstrated that the variable-length probabilistic rules are sufficiently effective for describing bi-directional transcription.

This paper discusses a methodology using a minimal set of sentences to adapt an existing TTS duration model to capture interspeaker variations. The assumption is that the original duration database contains information of both language-specific and speaker-specific duration characteristics. In training a duration model for a new speaker, only the speaker-specific information needs to be modeled, therefore the size of the training data can be reduced drastically. Results from several experiments are compared and discussed.