Improving speech-to-text for the Indonesian language using a modified transformer

Analyzing teachers’ discourse plays a fundamental role in educational research and is a key component of Teaching Analytics. This usually involves transcribing lessons from audio recordings. As the number of recordings grows, Automatic Speech Recognition (ASR) systems gain popularity as a means for transcribing these recordings. However, most ASR systems are trained over very specific domains which usually involve read text and low environmental noise. This suggests common ASR systems available on the market may underperform over classroom recordings, as they present a unique type of environmental sound and spontaneous discourse, as opposed to the usual training domains. To address this challenge we present a system that automatically transcribes classroom discourse in a robust way with regard to classroom noise, which was trained over few annotated data. In particular, we used a state-of-the-art ASR model based on wav2vec 2.0 and fine-tuned it over a 6-h dataset of 4th to 8th grade Chilean lessons. We found that by leveraging its transformer-based architecture and changing the fine-tuning domain to classroom recordings, we can obtain a more accurate and robust transcriber for this source of audio which outperforms other popular cloud-based systems up to 35% and 59% in terms of Word and Character Error Rates, respectively. This work contributes by using state-of-the-art ASR techniques to develop a tool which is particularly adapted to classroom environments, making it robust and more reliable with regard to their environmental sound and the way teaching discourse is carried out.KeywordsAutomatic speech recognitionClassroom discourseTeaching analyticsTransfer learning

Development of a code-switched Hindi-Marathi dataset and transformer-based architecture for enhanced speech recognition using dynamic switching algorithms

Modeling the speaker variability is a key challenge for automatic speech recognition (ASR) systems. In this paper, the learning hidden unit contributions (LHUC) based adaptation techniques with compact speaker dependent (SD) parameters are used to facilitate both speaker adaptive training (SAT) and unsupervised test-time speaker adaptation for end-to-end (E2E) lattice-free MMI (LF-MMI) models. An unsupervised model-based adaptation framework is proposed to estimate the SD parameters in E2E paradigm using LF-MMI and cross entropy (CE) criterions. Various regularization methods of the standard LHUC adaptation, e.g., the Bayesian LHUC (BLHUC) adaptation, are systematically investigated to mitigate the risk of overfitting, on E2E LF-MMI CNN-TDNN and CNN-TDNN- BLSTM models. Lattice-based confidence score estimation is used for adaptation data selection to reduce the supervision label uncertainty. Experiments on the 300-hour Switchboard task suggest that, applying BLHUC in the proposed unsupervised E2E adaptation framework to byte pair encoding (BPE) based E2E LF-MMI systems consistently outperformed the baseline systems by relative word error rate (WER) reductions up to 10.5% and 14.7% on the NIST Hub5’00 and RT03 evaluation sets, and achieved the best performance in WERs of 9.0% and 9.7%, respectively. These results are comparable to the results of state-of-the-art adapted LF-MMI hybrid systems and adapted Conformer-based E2E systems.

This study explores the feasibility of constructing a small-scale speech recognition system capable of competing with larger, modern automated speech recognition (ASR) systems in both performance and word error rate (WER). Our central hypothesis posits that a compact transformer-based ASR model can yield comparable results, specifically in terms of WER, to traditional ASR models while challenging contemporary ASR systems that boast significantly larger computational sizes. The aim is to extend ASR capabilities to under-resourced languages with limited corpora, catering to scenarios where practitioners face constraints in both data availability and computational resources. The model, comprising a compact convolutional neural network (CNN) and transformer architecture with 2.214 million parameters, challenges the conventional wisdom that large-scale transformer-based ASR systems are essential for achieving high accuracy. In comparison, contemporary ASR systems often deploy over 300 million parameters. Trained on a modest dataset of approximately 3000 h – significantly less than the 50,000 h used in larger systems – the proposed model leverages the Common Voice and LibriSpeech datasets. Evaluation on the LibriSpeech test-clean and test-other datasets produced character error rates (CERs) of 6.40% and 16.73% and WERs of 16.03% and 35.51% respectively. Comparisons with existing architectures showcase the efficiency of our model. A gated recurrent unit (GRU) architecture, albeit achieving lower error rates, incurred a computational cost 24 times larger than our proposed model. Large-scale transformer architectures, while achieving marginally lower WERs (2%–4% on LibriSpeech test-clean), require 200 times more parameters and 53,000 additional hours of training data. Modern large language models are used to improve the WERs, but require large computational resources. To further enhance performance, a small 4-g language model was integrated into our end-to-end ASR model, resulting in improved WERs. The overarching goal of this work is to provide a practical solution for practitioners dealing with limited datasets and computational resources, particularly in the context of under-resourced languages.

State-of-the-art multilingual Automatic Speech Recognition (ASR) models produce systematic errors when applied to low-resource languages like Rajasthani, for which they lack dedicated training data. This article addresses this challenge by introducing a post-ASR correction framework that leverages the complementary error patterns in the outputs (termed as views) from two distinct models: Whisper-large-v3 and MMS-1B-All. We propose a multi-view, 1 character-level sequence-to-sequence (Seq2Seq) model that uses a gated fusion mechanism to dynamically weigh information from the two ASR outputs. On a new benchmark created from the IndicTTS Rajasthani corpus, our gated model achieves a Character Error Rate (CER) of 7.86% and a Word Error Rate (WER) of 29.98%. This outperforms the best single-view baselines (8.01% CER and 30.33% WER), simple multi-view concatenation (8.21% CER and 30.05% WER), as well as Llama-3.2-3B and mBART-50-large, both fine-tuned on Whisper and MMS inputs. It also surpasses powerful Large Language Models (LLMs) like GPT-4o and Gemini 2.5 Pro in a zero-shot setting. This work establishes the first baseline for post-ASR correction in Rajasthani, demonstrating that a compact, specialized model is more effective than general-purpose LLMs for this targeted, low-resource task.

This work presents a dialect-aware and noise-robust Kannada automatic speech recognition (ASR) system that bridges the gap between low-resource linguistic contexts and state-of-the-art deep learning models. We design a two-stage approach: (i) a scratch-built convolutional neural network (CNN)–Transformer hybrid trained on curated Kannada speech data with fast Fourier transform-based noise reduction and (ii) fine-tuning OpenAI’s Whisper-small model on a dialect-diverse corpus. The proposed pipeline integrates adaptive noise suppression, subword tokenization, and beam-search decoding to handle agglutinative morphology, speaker variation, and environmental noise. Extensive experiments were conducted on two datasets: the Few-shot Learning Evaluation of Universal Representations of Speech (FLEURS) multilingual Kannada subset and a curated 150-sample, Custom-collected Kannada speech dataset covering both formal and conversational speech. On the FLEURS test set (≈ 2.5 hours), our fine-tuned Whisper model achieves a word error rate (WER) of 0.15, a character error rate (CER) of 0.255, and a BLEU score of 0.912, representing a 53% relative reduction in WER and 36% reduction in CER compared to the scratch CNN–Transformer baseline. For the customized dataset, the fine-tuned Whisper model achieves a WER of 0.2311 and a CER of 0.0453, outperforming Google Speech-to-Text Application Programming Interface (API) by 16.8% (relative WER reduction) and surpassing the scratch transformer by over 70% in WER. We further evaluate robustness under dialectal variation and noisy recordings, providing detailed error analysis and computational efficiency metrics. To our knowledge, this is the first comprehensive evaluation of Whisper fine-tuning for Kannada, demonstrating its viability for real-time, edge-deployable applications in education, accessibility, and public administration.

In this paper, the improvements in the recently implemented Kannada speech recognition system is demonstrated in detail. The Kannada automatic speech recognition (ASR) system consists of ASR models which are created by using Kaldi, IVRS call flow and weather and agricultural commodity prices information databases. The task specific speech data used in the recently developed spoken dialogue system had high level of different background noises. The different types of noises present in collected speech data had an adverse effect on the on line and off line speech recognition performances. Therefore, to improve the speech recognition accuracy in Kannada ASR system, a noise reduction algorithm is developed which is a fusion of spectral subtraction with voice activity detection (SS-VAD) and minimum mean square error spectrum power estimator based on zero crossing (MMSE-SPZC) estimator. The noise elimination algorithm is added in the system before the feature extraction part. An alternative ASR models are created using subspace Gaussian mixture models (SGMM) and deep neural network (DNN) modeling techniques. The experimental results show that, the fusion of noise elimination technique and SGMM/DNN based modeling gives a better relative improvement of 7.68% accuracy compared to the recently developed GMM-HMM based ASR system. The least word error rate (WER) acoustic models could be used in spoken dialogue system. The developed spoken query system is tested from Karnataka farmers under uncontrolled environment.

We report insights from translating Spanish conversational telephone speech into English text by cascading an automatic speech recognition (ASR) system with a statistical machine translation (SMT) system. The key new insight is that the informal register of conversational speech is a greater challenge for ASR than for SMT: the BLEU score for translating the reference transcript is 64%, but drops to 32% for translating automatic transcripts, whose word error rate (WER) is 40%. Several strategies are examined to mitigate the impact of ASR errors on the SMT output: (i) providing the ASR lattice, instead of the 1-best output, as input to the SMT system, (ii) training the SMT system on Spanish ASR output paired with English text, instead of Spanish reference transcripts, and (iii) improving the core ASR system. Each leads to consistent and complementary improvements in the SMT output. Compared to translating the 1-best output of an ASR system with 40% WER using an SMT system trained on Spanish reference transcripts, translating the output lattice of a better ASR system with 35% WER using an SMT system trained on ASR output improves BLEU from 32% to 38%.

A deep neural network (DNN)-based speech enhancement (SE) aiming to maximize the performance of an automatic speech recognition (ASR) system is proposed in this paper. In order to optimize the DNN-based SE model in terms of the character error rate (CER), which is one of the metric to evaluate the ASR system and generally non-differentiable, our method uses two DNNs: one for speech processing and one for mimicking the output CERs derived through an acoustic model (AM). Then both of DNNs are alternately optimized in the training phase. Even if the AM is a black-box, e.g., like one provided by a third-party, the proposed method enables the DNN-based SE model to be optimized in terms of the CER since the DNN mimicking the AM is differentiable. Consequently, it becomes feasible to build CER-centric SE model that has no negative effect, e.g., additional calculation cost and changing network architecture, on the inference phase since our method is merely a training scheme for the existing DNN-based methods. Experimental results show that our method improved CER by 8.8% relative derived through a black-box AM although certain noise levels are kept.

Venous thromboembolism (VTE) is a common vascular disorder requiring extended anticoagulation therapy postdischarge to reduce recurrence risk. Home rehabilitation management systems that use electronic health records from hospital care provide opportunities for continuous patient monitoring. However, transferring medical data from clinical to home settings raises significant concerns about privacy and security. Conventional methods such as manual data entry, optical character recognition, and dedicated data transmission lines face notable technical and operational challenges. This study aims to develop a QR code-based security transmission algorithm using Avro and byte pair encoding (BPE). The algorithm supports the secure creation and transfer of out-of-hospital health records by enabling patients to scan QR codes via a dedicated mobile app, ensuring data security and user privacy. Between January and October 2024, 300 hospitalized patients with VTE were recruited at the Sixth Medical Center of the Chinese PLA General Hospital. Post discharge, participants used a home rehabilitation app tailored for VTE management. The QR code-based security transmission algorithm was developed to securely transfer in-hospital electronic health records to the out-of-hospital app. It uses BPE, Avro, and Gzip for optimized data compression and uses ChaCha20 and BLAKE3 for encryption and authentication. Specifically, BPE tokenizes medical text, while Avro serializes JSON (JavaScript Object Notation) objects, contributing to data encryption. A proprietary tokenizer was trained, and compression efficiency was evaluated using a "Performance Benchmark Dataset." Comparative analyses were conducted to assess the compression efficiency of JSON serialization methods (Avro and ASN.1 [Abstract Syntax Notation One]), and tokenization algorithms (BPE and unigram). The dataset consisted of JSON files from 300 patients, averaging 240.1 fields per file (range 89-623) and 7095 bytes in size (range 2748-17,425 bytes). Using the BPE + Avro + Gzip algorithm, the average file size was reduced to 1048 bytes, achieving a compression ratio of 6.67. This was 1.82 times more efficient than traditional Gzip compression (average file size: 1907 bytes; compression ratio: 3.66; P<.001). For Chinese medical text tokenization, BPE outperformed unigram with a compression ratio of 4.68 versus 4.55 (P<.001). Avro and ASN.1 demonstrated comparable compression ratios of 2.57 and 2.59, respectively, when used alone (P=.30). However, Avro combined with BPE and Gzip significantly outperformed ASN.1, achieving compression ratios of 6.67 versus 5.21 (P<.001). Additionally, 84.7% (254/300) of patients needed to scan only 1 QR code, requiring an average of 3.1 seconds. The QR code-based security transmission algorithm using Avro and BPE efficiently compresses and transmits data in an encrypted manner and authenticates the identity of the scanning users, ensuring the privacy and security of medical data. Delivered as a software development kit, the algorithm offers straightforward implementation and usability, supporting its broad adoption across various applications.

Speech foundation models trained at a massive scale, both in terms of model and data size, result in robust systems capable of performing multiple speech tasks, including automatic speech recognition (ASR).These models transcend language and domain barriers, yet effectively measuring their performance remains a challenge.Traditional metrics like word error rate (WER) and character error rate (CER) are commonly used to evaluate ASR performance but often fail to reflect transcription quality in critical contexts, particularly when detecting fabricated outputs.This phenomenon, known as hallucination, is especially concerning in high-stakes domains such as healthcare, legal, and aviation, where errors can have severe consequences.In our work, we address this gap by investigating hallucination in ASR models.We examine how factors such as distribution shifts, model size, and model architecture influence the hallucination error rate (HER), a metric we introduce to quantify hallucinations.Our analysis of over 20 ASR models reveals three key insights: (1) High WERs can mask low hallucination rates, while low WERs may conceal dangerous hallucinations.(2) Synthetic noise, both adversarial and common perturbations like white noise, pitch shift, and time stretching, increase HER.(3) Distribution shift correlates strongly with HER ( = 0.91).Our findings highlight the importance of incorporating HER alongside traditional metrics like WER to better assess ASR model performance, particularly in high-stakes domains.

A common metric for evaluating Automatic Speech Recognition (ASR) is Word Error Rate (WER) which solely takes into account discrepancies at the word-level. Although WER is useful, it is not guaranteed to correlate well with intelligibility or performance on downstream tasks that make use of ASR. Meaningful assess- ment of ASR mistakes becomes even more important in high-stake scenarios such as health-care. I propose 2 general measures to evaluate the quality or severity of mistakes made by ASR systems, one based on sentiment analysis and another based on text embeddings. Both have the potential to overcome the limitations of WER. I evaluate these measures on simulated patient-doctor conversations. Measures of severity based on sentiment ratings and text embeddings correlate with human ratings of severity. Measures based on text embeddings have the capability to predict human severity ratings better than WER. These measures are used in metrics in the overall evaluation of 5 ASR engines alongside WER. Results show that these metrics capture characteristics of ASR errors that WER does not. Furthermore, I train an ASR system using severity as a penalty in the loss function and demonstrate the potential for using severity not only in the evaluation, but in the development of ASR. Advantages and limitations of this methodology are analyzed and discussed.

Vocal Tract Length Normalization (VTLN) is used to design vocal tract length normalized Automatic Speech Recognition (ASR) systems. It has led to improvement in the performance of ASR systems by taking into account the physiological differences among speakers. Recently, a number of speech recognition applications are being developed for Indian languages. In this paper, we use state-of-the-art method for VTLN based on maximum likelihood approach. A vowel recognition system has been developed for two low resourced Indian languages, viz., Gujarati and Marathi. Appropriate warping factors have been obtained for all speakers considered for training and testing procedures. An improvement in the performance of vowel recognition is observed as compared to state-of-the-art Mel Frequency Cepstral Coefficients (MFCC).

This paper presents a novel multistage fine-tuning strategy designed to enhance automatic speech recognition (ASR) performance in low-resource languages using OpenAI’s Whisper model. In this approach we aim to build ASR model for languages with limited digital resources by sequentially adapting the model across linguistically similar languages. We experimented this on the Malasar language, a Dravidian language spoken by approximately ten thousand people in the Western Ghats of South India. The development of speech recognition for Malasar language faces significant barriers due to its lack of written form and the absence of digital speech datasets needed for training. Working in collaboration with Wycliffe India and Malasar community members, we created a spoken Malasar corpus paired with transcription in Tamil script, a closely related major language. In our approach to build ASR model for Malasar, we first built an intermediate Tamil ASR, leveraging higher data availability for Tamil annotated speech. This intermediate model is subsequently fine-tuned on Malasar data, allowing for more effective ASR adaptation despite limited resources. The multistage fine-tuning strategy demonstrated significant improvements over direct fine-tuning on Malasar data alone, achieving a word error rate (WER) of 51.9%, which is 4.5% absolute reduction when compared to the direct fine-tuning method. Further a WER reduction to 47.3% was achieved through punctuation removal in post-processing, which addresses formatting inconsistencies that impact evaluation. Our results underscore the effectiveness of sequential multistage fine-tuning combined with targeted post-processing as a scalable strategy for ASR system development in low-resource languages, especially where linguistic similarities can be leveraged to bridge gaps in training data.

This article concerns comparative studies on the Automatic Speech Recognition (ASR) model incorporated with the Large Language Model (LLM) used for medical interviews. The proposed solution is tested on polish language benchmarks and dataset with medical interviews. The latest ASR technologies are based on convolutional neural networks (CNNs), recurrent neural networks (RNNs) and Transformers. Most of them work as end-to-end solutions. The presented approach in the case of the Whisper model shows a two-stage solution with End-To-End ASR and LLM working together in a pipeline. The ASR output is an input for LLM. The LLM is a component by which the output from ASR is corrected and improved. Comparative studies for automatic recognition of the Polish language between modern End-To-End deep learning architectures and the ASR hybrid model were performed. The medical interview tests were performed with two state-of-the-art ASR models: OpenAI Whisper incorporated with LLM and Scribe ElevenLabs. Additionally, the results were compared with five more end-to-end models (QuartzNet, FastConformer, Wav2Vec 2.0 XLSR and ESPnet Model Zoo) on Mozilla Common Voice and VoxPopuli databases. Tests were conducted for clean audio signal, signal with bandwidth limitation, and degraded. The tested models were evaluated on the basis of Word Error Rate (WER) and Character Error Rate (CER). The results show that the Whisper model performs by far the best among the open-source models. ElevenLabs Scribe model, on the other hand, performs best for Polish on both general benchmark and medical data.