Tanel Alumäe

Tanel Alumäe, Kunnar Kukk

This paper investigates different pretraining approaches to spoken language identification. The paper is based on our submission to the Oriental Language Recognition 2021 Challenge. We participated in two tracks of the challenge: constrained and unconstrained language recognition. For the constrained track, we first trained a Conformer-based encoder-decoder model for multilingual automatic speech recognition (ASR), using the provided training data that had transcripts available. The shared encoder of the multilingual ASR model was then finetuned for the language identification task. For the unconstrained task, we relied on both externally available pretrained models as well as external data: the multilingual XLSR-53 wav2vec2.0 model was finetuned on the VoxLingua107 corpus for the language recognition task, and finally finetuned on the provided target language training data, augmented with CommonVoice data. Our primary metric $C_{\rm avg}$ values on the Test set are 0.0079 for the constrained task and 0.0119 for the unconstrained task which resulted in the second place in both rankings. In post-evaluation experiments, we study the amount of target language data needed for training an accurate backend model, the importance of multilingual pretraining data, and compare different models as finetuning starting points.

Joonas Kalda, Tanel Alumäe

In this paper, we present a novel training method for speaker change detection models. Speaker change detection is often viewed as a binary sequence labelling problem. The main challenges with this approach are the vagueness of annotated change points caused by the silences between speaker turns and imbalanced data due to the majority of frames not including a speaker change. Conventional training methods tackle these by artificially increasing the proportion of positive labels in the training data. Instead, the proposed method uses an objective function which encourages the model to predict a single positive label within a specified collar. This is done by marginalizing over all possible subsequences that have exactly one positive label within the collar. Experiments on English and Estonian datasets show large improvements over the conventional training method. Additionally, the model outputs have peaks concentrated to a single frame, removing the need for post-processing to find the exact predicted change point which is particularly useful for streaming applications.

Tanel Alumäe, Jiaming Kong, Daniil Robnikov

This paper describes Tallinn University of Technology (TalTech) systems developed for the ASRU MADASR 2023 Challenge. The challenge focuses on automatic speech recognition of dialect-rich Indian languages with limited training audio and text data. TalTech participated in two tracks of the challenge: Track 1 that allowed using only the provided training data and Track 3 which allowed using additional audio data. In both tracks, we relied on wav2vec2.0 models. Our methodology diverges from the traditional procedure of finetuning pretrained wav2vec2.0 models in two key points: firstly, through the implementation of the aligned data augmentation technique to enhance the linguistic diversity of the training data, and secondly, via the application of deep prefix tuning for dialect adaptation of wav2vec2.0 models. In both tracks, our approach yielded significant improvements over the provided baselines, achieving the lowest word error rates across all participating teams.

Tiia Sildam, Andra Velve, Tanel Alumäe

This paper investigates the finetuning of end-to-end models for bidirectional Estonian-English and Estonian-Russian conversational speech-to-text translation. Due to the limited availability of speech translation data for Estonian, we created additional training data by web scraping and synthesizing data from speech recognition datasets using machine translation. We evaluated three publicly available end-to-end models: Whisper, OWSM 3.1, and SeamlessM4T. Our results indicate that fine-tuning with synthetic data enhances translation accuracy by a large margin, with SeamlessM4T matching or surpassing cascaded speech translation systems that use state-of-the-art speech recognition and machine translation models.

Sandipana Dowerah, Atharva Kulkarni, Ajinkya Kulkarni et al.

Parallel to the development of advanced deepfake audio generation, audio deepfake detection has also seen significant progress. However, a standardized and comprehensive benchmark is still missing. To address this, we introduce Speech DeepFake (DF) Arena, the first comprehensive benchmark for audio deepfake detection. Speech DF Arena provides a toolkit to uniformly evaluate detection systems, currently across 14 diverse datasets and attack scenarios, standardized evaluation metrics and protocols for reproducibility and transparency. It also includes a leaderboard to compare and rank the systems to help researchers and developers enhance their reliability and robustness. We include 14 evaluation sets, 12 state-of-the-art open-source and 3 proprietary detection systems. Our study presents many systems exhibiting high EER in out-of-domain scenarios, highlighting the need for extensive cross-domain evaluation. The leaderboard is hosted on Huggingface1 and a toolkit for reproducing results across the listed datasets is available on GitHub.

Aleksei Dorkin, Taido Purason, Emil Kalbaliyev et al.

Large language models (LLMs) are predominantly trained on English-centric data, resulting in uneven performance for smaller languages. We study whether continued pretraining (CPT) can substantially improve Estonian capabilities in a pretrained multilingual LLM while preserving its English and general reasoning performance. Using Llama 3.1 8B as the main base model, we perform CPT on a mixture that increases Estonian exposure while approximating the original training distribution through English replay and the inclusion of code, mathematics, and instruction-like data. We subsequently apply supervised fine-tuning, preference optimization, and chat vector merging to introduce robust instruction-following behavior. Evaluation on a comprehensive suite of Estonian benchmarks shows consistent gains in linguistic competence, knowledge, reasoning, translation quality, and instruction-following compared to the original base model and its instruction-tuned variant, while maintaining competitive performance on English benchmarks. These findings indicate that CPT, with an appropriately balanced data mixture, together with post-training alignment, can substantially improve single-language capabilities in pretrained multilingual LLMs.

Helena Grete Lillepalu, Tanel Alumäe

The availability of LLM benchmarks for the Estonian language is limited, and a comprehensive evaluation comparing the performance of different LLMs on Estonian tasks has yet to be conducted. We introduce a new benchmark for evaluating LLMs in Estonian, based on seven diverse datasets. These datasets assess general and domain-specific knowledge, understanding of Estonian grammar and vocabulary, summarization abilities, contextual comprehension, and more. The datasets are all generated from native Estonian sources without using machine translation. We compare the performance of base models, instruction-tuned open-source models, and commercial models. Our evaluation includes 6 base models and 26 instruction-tuned models. To assess the results, we employ both human evaluation and LLM-as-a-judge methods. Human evaluation scores showed moderate to high correlation with benchmark evaluations, depending on the dataset. Claude 3.7 Sonnet, used as an LLM judge, demonstrated strong alignment with human ratings, indicating that top-performing LLMs can effectively support the evaluation of Estonian-language models.

Ajinkya Kulkarni, Sandipana Dowerah, Atharva Kulkarni et al.

Self-supervised learning (SSL) underpins modern audio deepfake detection, yet most prior work centers on a single large wav2vec2-XLSR backbone, leaving compact under studied. We present RAPTOR, Representation Aware Pairwise-gated Transformer for Out-of-domain Recognition a controlled study of compact SSL backbones from the HuBERT and WavLM within a unified pairwise-gated fusion detector, evaluated across 14 cross-domain benchmarks. We show that multilingual HuBERT pre-training is the primary driver of cross-domain robustness, enabling 100M models to match larger and commercial systems. Beyond EER, we introduce a test-time augmentation protocol with perturbation-based aleatoric uncertainty to expose calibration differences invisible to standard metrics: WavLM variants exhibit overconfident miscalibration under perturbation, whereas iterative mHuBERT remains stable. These findings indicate that SSL pre-training trajectory, not model scale, drives reliable audio deepfake detection.

Aleksei Žavoronkov, Tanel Alumäe

This paper presents an analysis of three end-to-end models developed for the NOCASA 2025 Challenge, aimed at automatic word-level pronunciation assessment for children learning Norwegian as a second language. Our models include an encoder-decoder Siamese architecture (E2E-R), a prefix-tuned direct classification model leveraging pretrained wav2vec2.0 representations, and a novel model integrating alignment-free goodness-of-pronunciation (GOP) features computed via CTC. We introduce a weighted ordinal cross-entropy loss tailored for optimizing metrics such as unweighted average recall and mean absolute error. Among the explored methods, our GOP-CTC-based model achieved the highest performance, substantially surpassing challenge baselines and attaining top leaderboard scores.

Tanel Alumäe, Artem Fedorchenko

This paper describes the language identification and multilingual speech recognition system developed at Tallinn University of Technology for the Interspeech 2025 ML-SUPERB 2.0 Challenge. A hybrid language identification system is used, consisting of a pretrained language embedding model and a light-weight speech recognition model with a shared encoder across languages and language-specific bigram language models. For speech recognition, three models are used, where only a single model is applied for each language, depending on the training data availability and performance on held-out data. The model set consists of a finetuned version of SeamlessM4T, MMS-1B-all with custom language adapters and MMS-zeroshot. The system obtained the top overall score in the challenge.

Artem Fedorchenko, Tanel Alumäe

This paper presents an approach for generating high-quality, same-language subtitles for Estonian TV content. We fine-tune the Whisper model on human-generated Estonian subtitles and enhance it with iterative pseudo-labeling and large language model (LLM) based post-editing. Our experiments demonstrate notable subtitle quality improvement through pseudo-labeling with an unlabeled dataset. We find that applying LLM-based editing at test time enhances subtitle accuracy, while its use during training does not yield further gains. This approach holds promise for creating subtitle quality close to human standard and could be extended to real-time applications.

Adrian Łańcucki, Jan Chorowski, Guillaume Sanchez et al.

In this paper we demonstrate methods for reliable and efficient training of discrete representation using Vector-Quantized Variational Auto-Encoder models (VQ-VAEs). Discrete latent variable models have been shown to learn nontrivial representations of speech, applicable to unsupervised voice conversion and reaching state-of-the-art performance on unit discovery tasks. For unsupervised representation learning, they became viable alternatives to continuous latent variable models such as the Variational Auto-Encoder (VAE). However, training deep discrete variable models is challenging, due to the inherent non-differentiability of the discretization operation. In this paper we focus on VQ-VAE, a state-of-the-art discrete bottleneck model shown to perform on par with its continuous counterparts. It quantizes encoder outputs with on-line $k$-means clustering. We show that the codebook learning can suffer from poor initialization and non-stationarity of clustered encoder outputs. We demonstrate that these can be successfully overcome by increasing the learning rate for the codebook and periodic date-dependent codeword re-initialization. As a result, we achieve more robust training across different tasks, and significantly increase the usage of latent codewords even for large codebooks. This has practical benefit, for instance, in unsupervised representation learning, where large codebooks may lead to disentanglement of latent representations.

Tanel Alumäe, Ottokar Tilk, Asadullah

This paper describes the current TTÜ speech transcription system for Estonian speech. The system is designed to handle semi-spontaneous speech, such as broadcast conversations, lecture recordings and interviews recorded in diverse acoustic conditions. The system is based on the Kaldi toolkit. Multi-condition training using background noise profiles extracted automatically from untranscribed data is used to improve the robustness of the system. Out-of-vocabulary words are recovered using a phoneme n-gram based decoding subgraph and a FST-based phoneme-to-grapheme model. The system achieves a word error rate of 8.1% on a test set of broadcast conversations. The system also performs punctuation recovery and speaker identification. Speaker identification models are trained using a recently proposed weakly supervised training method.

Martin Karu, Tanel Alumäe

We propose an approach for training speaker identification models in a weakly supervised manner. We concentrate on the setting where the training data consists of a set of audio recordings and the speaker annotation is provided only at the recording level. The method uses speaker diarization to find unique speakers in each recording, and i-vectors to project the speech of each speaker to a fixed-dimensional vector. A neural network is then trained to map i-vectors to speakers, using a special objective function that allows to optimize the model using recording-level speaker labels. We report experiments on two different real-world datasets. On the VoxCeleb dataset, the method provides 94.6% accuracy on a closed set speaker identification task, surpassing the baseline performance by a large margin. On an Estonian broadcast news dataset, the method provides 66% time-weighted speaker identification recall at 93% precision.

Ottokar Tilk, Tanel Alumäe

Recent neural headline generation models have shown great results, but are generally trained on very large datasets. We focus our efforts on improving headline quality on smaller datasets by the means of pretraining. We propose new methods that enable pre-training all the parameters of the model and utilize all available text, resulting in improvements by up to 32.4% relative in perplexity and 2.84 points in ROUGE.