Audio comprehension-including speech, non-speech sounds, and music-is essential for achieving human-level intelligence. Consequently, AI agents must demonstrate holistic audio understanding to qualify as generally intelligent. However, evaluating auditory intelligence comprehensively remains challenging. To address this gap, we introduce MMAU-Pro, the most comprehensive and rigorously curated benchmark for assessing audio intelligence in AI systems. MMAU-Pro contains 5,305 instances, where each instance has one or more audios paired with human expert-generated question-answer pairs, spanning speech, sound, music, and their combinations. Unlike existing benchmarks, MMAU-Pro evaluates auditory intelligence across 49 unique skills and multiple complex dimensions, including long-form audio comprehension, spatial audio reasoning, multi-audio understanding, among others. All questions are meticulously designed to require deliberate multi-hop reasoning, including both multiple-choice and open-ended response formats. Importantly, audio data is sourced directly ``from the wild" rather than from existing datasets with known distributions. We evaluate 22 leading open-source and proprietary multimodal AI models, revealing significant limitations: even state-of-the-art models such as Gemini 2.5 Flash and Audio Flamingo 3 achieve only 59.2% and 51.7% accuracy, respectively, approaching random performance in multiple categories. Our extensive analysis highlights specific shortcomings and provides novel insights, offering actionable perspectives for the community to enhance future AI systems' progression toward audio general intelligence. The benchmark and code is available at https://sonalkum.github.io/mmau-pro.

We present MADASR 2.0, a challenge at ASRU 2025 aimed at advancing multilingual and multidialectal automatic speech recognition (ASR) in low-resource Indian languages. Building on the 2023 edition, it introduces a subset of the RESPIN corpus, over 1200 hours of read speech across 8 languages and 33 dialects, with test sets including both read and spontaneous speech. The challenge comprises four tracks varying by training data size and external resource usage, and supports auxiliary tasks like language and dialect identification. We detail the dataset, tasks, baselines, and submissions and analyse trends across tracks and speech styles. Results highlight the continued difficulty of spontaneous ASR, the benefits of multitask and transfer learning, and effective strategies for building dialect-aware ASR systems. MADASR 2.0 offers a standardised benchmark to support future research on inclusive and scalable ASR for linguistically diverse populations.

Accurate, low-latency endpointing is crucial for effective spoken dialogue systems. While traditional endpointers often rely on spectrum-based audio features, this work proposes real-time speech endpointing for multi-turn dialogues using streaming, low-bitrate Neural Audio Codec (NAC) features, building upon recent advancements in neural audio codecs. To further reduce cutoff errors, we introduce a novel label delay training scheme. At a fixed median latency of 160 ms, our combined NAC and label delay approach achieves significant relative cutoff error reductions: 42.7% for a single-stream endpointer and 37.5% for a two-stream configuration, compared to baseline methods. Finally, we demonstrate efficient integration with a codec-based pretrained speech large language model, improving its median response time by 1200 ms and reducing its cutoff error by 35%.

Dialect identification (DID) addresses the challenge of recog-nizing regional variations within a language. The current deep learning approaches focus on audio-only, text-only, or multi-task setups combining automatic speech recognition (ASR) with DID. This work introduces a novel multimodal architecture that leverages speech and text features to enhance DID performance. Our method integrates ASR-generated speech representations with text embeddings derived from ASR hypotheses using a RoBERTa-based encoder. Additionally, we perform a layer-wise analysis of the IndicWav2Vec model to identify the layers most effective for extracting dialectal features. We evaluate our approach on a subset of the RESPIN dataset featuring eight Indian languages and 33 dialects. Experimental results show that our proposed multimodal DID system achieves an average DID accuracy of 79.81%, consistently outperforming baseline methods. This study is the first to analyse comprehensively DID in Indian languages, providing new insights into their dialectal diversity.

We introduce RESPIN-S1.0, the largest publicly available dialect-rich read-speech corpus for Indian languages, comprising more than 10,000 hours of validated audio across nine major languages: Bengali, Bhojpuri, Chhattisgarhi, Hindi, Kannada, Magahi, Maithili, Marathi, and Telugu. Indian languages exhibit high dialectal variation and are spoken by populations that remain digitally underserved. Existing speech corpora typically represent only standard dialects and lack domain and linguistic diversity. RESPIN-S1.0 addresses this limitation by collecting speech across more than 38 dialects and two high-impact domains: agriculture and finance. Text data were composed by native dialect speakers and validated through a pipeline combining automated and manual checks. Over 200,000 unique sentences were recorded through a crowdsourced mobile platform and categorised into clean, semi-noisy, and noisy subsets based on transcription quality, with the clean portion alone exceeding 10,000 hours. Along with audio and transcriptions, RESPIN provides dialect-aware phonetic lexicons, speaker metadata, and reproducible train, development, and test splits. To benchmark performance, we evaluate multiple ASR models, including TDNN-HMM, E-Branchformer, Whisper, and wav2vec2-based self-supervised models, and find that fine-tuning on RESPIN significantly improves recognition accuracy over pretrained baselines. A subset of RESPIN-S1.0 has already supported community challenges such as the SLT Code Hackathon 2022 and MADASR@ASRU 2023 and 2025, releasing more than 1,200 hours publicly. This resource supports research in dialectal ASR, language identification, and related speech technologies, establishing a comprehensive benchmark for inclusive, dialect-rich ASR in multilingual low-resource settings. Dataset: https://spiredatasets.ee.iisc.ac.in/respincorpus Code: https://github.com/labspire/respin_baselines.git

In this work, several methods have been proposed towards improving the performance of dialectal automatic speech recognition (ASR). A novel encoder architecture has been introduced that is suited for multi-dialect ASR training. Further, we propose Multi-Task Self-Supervised learning (SSL) fine-tuning using CTC and dialect identification. Additionally, the use of different language models (LM) to improve the performance of dialectal ASR has been investigated. Around 800 hours of Bengali and Bhojpuri data, released as a part of the MADASR ASRU challenge have been used to train these models. The work shows that the proposed multi-encoder ASR observes a relative reduction of 7.5% and 9% in WER in Bhojpuri and Bengali, respectively. Additionally, we also observe a 1-2% WER reduction in fine-tuning SSL, further improving performance in these languages. Moreover, we observe advantages in using dialect-specific LM decoding based on predicted dialect.

The Lightweight, Multi-speaker, Multi-lingual Indic Text-to-Speech (LIMMITS’23) challenge is organized as part of the ICASSP 2023 signal processing grand challenge. LIMMITS’23 aims at the development of a lightweight, multi-speaker, multi-lingual Text to Speech (TTS) model using datasets in Marathi, Hindi, and Telugu. The challenge encourages the advancement of TTS in Indian Languages as well as the development of techniques involved in TTS data selection and model compression. The 3 tracks of LIMMITS’23 have provided an opportunity for various researchers and practitioners around the world to explore the state of the art in TTS research.

Real-Time Magnetic resonance imaging (rtMRI) of the midsagittal plane of the mouth is of interest for speech production research. In this work, we focus on estimating utterance level rtMRI video from the spoken phoneme sequence. We obtain time-aligned phonemes from forced alignment, to obtain frame-level phoneme sequences which are aligned with rtMRI frames. We propose a sequence-tosequence learning model with a transformer phoneme encoder and convolutional frame decoder. We then modify the learning by using intermediary features obtained from sampling from a pretrained phoneme-conditioned variational autoencoder (CVAE). We train on 8 subjects in a subject-specific manner and demonstrate the performance with a subjective test. We also use an auxiliary task of air tissue boundary (ATB) segmentation to obtain the objective scores on the proposed models. We show that the proposed method is able to generate realistic rtMRI video for unseen utterances, and adding CVAE is beneficial for learning the sequence-to-sequence mapping for subjects where the mapping is hard to learn.

In this work, we investigate the effectiveness of pretrained Self-Supervised Learning (SSL) features for learning the mapping for acoustic to articulatory inversion (AAI). Signal processing-based acoustic features such as MFCCs have been predominantly used for the AAI task with deep neural networks. With SSL features working well for various other speech tasks such as speech recognition, emotion classification, etc., we experiment with its efficacy for AAI. We train on SSL features with transformer neural networks-based AAI models of 3 different model complexities and compare its performance with MFCCs in subject-specific (SS), pooled and fine-tuned (FT) configurations with data from 10 subjects, and evaluate with correlation coefficient (CC) score on the unseen sentence test set. We find that acoustic feature reconstruction objective-based SSL features such as TERA and DeCoAR work well for AAI, with SS CCs of these SSL features reaching close to the best FT CCs of MFCC. We also find the results consistent across different model sizes.

This paper describes the corpus and baseline systems for the Gram Vaani Automatic Speech Recognition (ASR) challenge in regional variations of Hindi. The corpus for this challenge comprises the spontaneous telephone speech recordings collected by a social technology enterprise, Gram Vaani. The regional variations of Hindi together with spontaneity of speech, natural background and transcriptions with variable accuracy due to crowdsourcing make it a unique corpus for ASR on spontaneous telephonic speech. Around, 1108 hours of real-world spontaneous speech recordings, including 1000 hours of unlabelled training data, 100 hours of labelled training data, 5 hours of development data and 3 hours of evaluation data, have been released as a part of the challenge. The efficacy of both training and test sets are validated on different ASR systems in both traditional time-delay neural network-hidden Markov model (TDNN-HMM) frameworks and fully-neural end-to-end (E2E) setup. The word error rate (WER) and character error rate (CER) on eval set for a TDNN model trained on 100 hours of labelled data are 29.7 and 15.1, respectively. While, in E2E setup, WER and CER on eval set for a conformer model trained on 100 hours of data are 32.9 and 19.0, respectively.

Estimating speech articulatory movements from speech acoustics is known as Acoustic to Articulatory Inversion (AAI). Recently, transformer-based AAI models have been shown to achieve state-of-art performance. However, in transformer networks, the attention is applied over the whole utterance, thereby needing to obtain the full utterance before the inference, which leads to high latency and is impractical for streaming AAI. To enable streaming during inference, evaluation could be performed on non-overlapping chucks instead of a full utterance. However, due to a mismatch of the attention receptive field during training and evaluation, there could be a drop in AAI performance. To overcome this scenario, in this work we perform experiments with different attention masks and use context from previous predictions during training. Experiments results revealed that using the random start mask attention with the context from previous predictions of transformer decoder performs better than the baseline results.

We estimate articulatory movements in speech production from different modalities - acoustics and phonemes. Acoustic-to articulatory inversion (AAI) is a sequence-to-sequence task. On the other hand, phoneme to articulatory (PTA) motion estimation faces a key challenge in reliably aligning the text and the articulatory movements. To address this challenge, we explore the use of a transformer architecture - FastSpeech, with explicit duration modelling to learn hard alignments between the phonemes and articulatory movements. We also train a transformer model on AAI. We use correlation coefficient (CC) and root mean squared error (rMSE) to assess the estimation performance in comparison to existing methods on both tasks. We observe 154%, 11.8% & 4.8% relative improvement in CC with subject-dependent, pooled and fine-tuning strategies, respectively, for PTA estimation. Additionally, on the AAI task, we obtain 1.5%, 3% and 3.1% relative gain in CC on the same setups compared to the state-of-the-art baseline. We further present the computational benefits of having transformer architecture as representation blocks.

In this paper, we consider 2-class and 3-class classification problems for classifying patients with Amyotrophic Lateral Sclerosis (ALS), Parkinson's Disease (PD), and Healthy Controls (HC) using a CNNLSTM network. Classification performance is examined for three different tasks, namely, Spontaneous speech (SPON), Diadochokinetic rate (DIDK) and Sustained phoneme production (PHON). Experiments are conducted using speech data recorded from 60 ALS, 60 PD, and 60 HC subjects. Classifications using SVM and DNN are considered as baseline schemes. Classification accuracy of ALS and HC (indicated by ALS/HC) using CNN-LSTM has shown an improvement of 10.40%, 4.22% and 0.08% for PHON, SPON and DIDK tasks, respectively over the best of the baseline schemes. Furthermore, the CNN-LSTM network achieves the highest PD/HC classification accuracy of 88.5% for the SPON task and the highest 3-class (ALS/PD/HC) classification accuracy of 85.24% for the DIDK task. Experiments using transfer learning at low resource training data show that data from ALS benefits PD/HC classification and vice-versa. Experiments with fine-tuning weights of 3-class (ALS/PD/HC) classifier for 2-class classification (PD/HC or ALS/HC) gives an absolute improvement of 2% classification accuracy in SPON task when compared with randomly initialized 2-class classifier.