Evaluating Pretrained General-Purpose Audio Representations for Music Genre Classification

This study investigates the use of self-supervised learning embeddings, particularly BYOL-A, in conjunction with a deep neural network classifier for Music Genre Classification. Our experiments demonstrate that BYOL-A embeddings outperform other pre-trained models, such as PANNs and VGGish, achieving an accuracy of 81.5% on the GTZAN dataset and 64.3% on FMA-Small. The proposed DNN classifier improved performance by 10-16% over linear classifiers. We explore the effects of contrastive and triplet loss and multitask training with optimized loss weights, achieving the highest accuracy. To address cross dataset challenges, we combined GTZAN and FMA-Small into a unified 18-class label space for joint training, resulting in slight performance drops on GTZAN but comparable results on FMA-Small. The scripts developed in this work are publicly available.

Timbre-Aware LLM-based Direct Speech-to-Speech Translation Extendable to Multiple Language Pairs

Direct Speech-to-Speech Translation (S2ST) has gained increasing attention for its ability to translate speech from one language to another, while reducing error propagation and latency inherent in traditional cascaded pipelines. However, existing direct S2ST systems continue to face notable challenges, including instability in semantic-acoustic alignment when parallel speech data is scarce, difficulty in preserving speaker identity, and limited multilingual scalability. In this work, we introduce DS2ST-LM, a scalable, single-stage direct S2ST framework leveraging a multilingual Large Language Model (LLM). The architecture integrates a Whisper speech encoder, a learnable projection module, a Qwen2-0.5B LLM, and a timbre-controlled vocoder. We construct GigaS2S-1000, a 1000-hour bilingual corpus by extending the GigaST dataset with high-fidelity synthetic target speech, and show that this synthetic data alleviates data scarcity to some extent. We investigate two semantic token generation strategies: speech-derived S3 tokens and text-derived tokens generated by a pre-trained LLM, and analyze their impact on training stability and semantic consistency. We further evaluate three projection architectures (Linear, Conv1D-Linear, and Q-Former) and observe that while higher-capacity projectors converge faster, the simple Linear projector achieves higher performance. Extensive experiments demonstrate that DS2ST-LM outperforms traditional cascaded and ST (Qwen-Audio) + TTS baselines across both lexical (BLEU, METEOR) and semantic (BLEURT, COMET) metrics, while extending to multiple language pairs, including French, Spanish, German, Hindi, Bengali, and Urdu. Furthermore, we incorporate timbre-aware speech synthesis to preserve speaker information, enabling DS2ST-LM to surpass prior direct S2ST systems in both speaker similarity and perceptual naturalness.

I-MSV 2022: Indic-Multilingual and Multi-sensor Speaker Verification Challenge

Speaker Verification (SV) is a task to verify the claimed identity of the claimant using his/her voice sample. Though there exists an ample amount of research in SV technologies, the development concerning a multilingual conversation is limited. In a country like India, almost all the speakers are polyglot in nature. Consequently, the development of a Multilingual SV (MSV) system on the data collected in the Indian scenario is more challenging. With this motivation, the Indic- Multilingual Speaker Verification (I-MSV) Challenge 2022 has been designed for understanding and comparing the state-of-the-art SV techniques. For the challenge, approximately $100$ hours of data spoken by $100$ speakers has been collected using $5$ different sensors in $13$ Indian languages. The data is divided into development, training, and testing sets and has been made publicly available for further research. The goal of this challenge is to make the SV system robust to language and sensor variations between enrollment and testing. In the challenge, participants were asked to develop the SV system in two scenarios, viz. constrained and unconstrained. The best system in the constrained and unconstrained scenario achieved a performance of $2.12\%$ and $0.26\%$ in terms of Equal Error Rate (EER), respectively.