823-OLT @ BUET DL Sprint 4.0: Context-Aware Windowing for ASR and Fine-Tuned Speaker Diarization in Bengali Long Form Audio

Ratnajit Dhar; Arpita Mallik

cs.SDFeb 24, 2026

823-OLT @ BUET DL Sprint 4.0: Context-Aware Windowing for ASR and Fine-Tuned Speaker Diarization in Bengali Long Form Audio

Authors: Ratnajit Dhar, Arpita Mallik

Organizations: Department of Computer Science and Engineering, Chittagong University of Engineering and Technology, Bangladesh

Abstract

Bengali, despite being one of the most widely spoken languages globally, remains underrepresented in long form speech technology, particularly in systems addressing transcription and speaker attribution. We present frameworks for long form Bengali speech intelligence that address automatic speech recognition using a Whisper Medium based model and speaker diarization using a finetuned segmentation model. The ASR pipeline incorporates vocal separation, voice activity detection, and a gap aware windowing strategy to construct context preserving segments for stable decoding. For diarization, a pretrained speaker segmentation model is finetuned on the official competition dataset (provided as part of the DL Sprint 4.0 competition organized under BUET CSE Fest), to better capture Bengali conversational patterns. The resulting systems deliver both efficient transcription of long form audio and speaker aware transcription to provide scalable speech technology solutions for low resource languages.

Figures & tables

Panel of Figure 1 — Fig. 1: Long-form Bengali ASR System Architecture.

Panel of Figure 2 — Fig. 2: Diarization System Architecture.

Hyperparameter	Value
Optimizer	AdamW
Learning Rate	$5\times 10^{-4}$
Batch Size	8
Gradient Accumulation	2 steps
Number of Epochs	12
Learning Rate Scheduler	Cosine Decay

TABLE I: Best-Performing Training Configuration

Demucs	VAD Threshold	Private	Public
Yes	Default	0.36240	0.34253
Yes	0.3	0.36491	0.34112
No	0.3	0.38920	0.37004
Yes	0.4	0.36507	0.34123
No	0.2	0.41841	0.37571

TABLE II: ASR results under different preprocessing configurations.

Epochs	Batch	Demucs	DER (Public / Private)
12	4	Yes	0.20249 / 0.29430
12	8	Yes	0.19284 / 0.29085
15	8	Yes	0.18758 / 0.29718
18	8	Yes	0.20406 / 0.28816
18	8	No	0.20292 / 0.28857

TABLE III: Fine-tuned diarization results with learning rate

5\times 10^{-4}

and warmup ratio

0.10

Clustering / Refinement Variant	Private DER	Public DER
PLDA scoring $\rightarrow$ spectral clustering	0.53437	0.55581
Spectral clustering + VBx refinement	0.42715	0.33578
DBSCAN clustering	0.67882	0.69779
Spectral clustering (no VBx)	0.42535	0.28062
Pyannote without Training	0.44885	0.40084

TABLE IV: Training-free Speaker Diarization Pipelines Scores.

Long-Form Speech Recognition

See all 21 papers

An Investigation Into Various Approaches For Bengali Long-Form Speech Transcription and Bengali Speaker Diarization
Mar 3, 2026Epshita Jahan, Khandoker Md Tanjinul Islam, Pritom Biswas +1Long-Form Speech Recognition Speaker Diarization
Department of Computer Science and Engineering, Bangladesh University of Engineering and Technology, Dhaka, Bangladesh
Robust Long-Form Bangla Speech Processing: Automatic Speech Recognition and Speaker Diarization
Feb 25, 2026MD. Sagor Chowdhury, Adiba Fairooz ChowdhuryLong-Form Speech Recognition Automatic Speech Recognition
Department of Computer Science and Engineering Chittagong University of Engineering and Technology Chattogram, Bangladesh
WhisperAlign: Word-Boundary-Aware ASR and WhisperX-Anchored Pyannote Diarization for Long-Form Bengali Speech
Mar 5, 2026Aurchi Chowdhury, Rubaiyat -E-Zaman, Sk. Ashrafuzzaman NafeesLong-Form Speech Recognition Speaker Diarization
CSE, BUET

Mar 3, 2026J/K move · Enter open · S save

An Investigation Into Various Approaches For Bengali Long-Form Speech Transcription and Bengali Speaker Diarization

Epshita Jahan, Khandoker Md Tanjinul Islam, Pritom Biswas, Tafsir Al Nafin

Department of Computer Science and Engineering, Bangladesh University of Engineering and Technology, Dhaka, Bangladesh

Long-Form Speech Recognition Speaker Diarization Automatic Speech Recognition Low-Resource Speech Recognition

Open paper AskPDF arXiv

Bengali remains a low-resource language in speech technology, especially for complex tasks like long-form transcription and speaker diarization. This paper presents a multistage approach developed for the "DL Sprint 4.0 - Bengali Long-Form Speech Recognition" and "DL Sprint 4.0 - Bengali Speaker Diarization" competitions on Kaggle, addressing the challenge of "who spoke when/what" in hour-long recordings. We implemented Whisper Medium fine-tuned on Bengali data (bengaliAI/tugstugi bengaliai-asr whisper-medium) for transcription and integrated pyannote/speaker-diarization-community-1 with our custom-trained segmentation model to handle diverse and noisy acoustic environments. Using a two-pass method with hyperparameter tuning, we achieved a DER of 0.27 on the private leaderboard and 0.19 on the public leaderboard. For transcription, chunking, background noise cleaning, and algorithmic post-processing yielded a WER of 0.38 on the private leaderboard. These results show that targeted tuning and strategic data utilization can significantly improve AI inclusivity for South Asian languages. All relevant code is available at: https://github.com/Short-Potatoes/Bengali-long-form-transcription-and-diarization.git Index Terms: Bengali speech recognition, speaker diarization, Whisper, ASR, low-resource languages, pyannote, voice activity detection

Abstract

Long-Form Speech Recognition

Explore similar work

An Investigation Into Various Approaches For Bengali Long-Form Speech Transcription and Bengali Speaker Diarization

Robust Long-Form Bangla Speech Processing: Automatic Speech Recognition and Speaker Diarization

WhisperAlign: Word-Boundary-Aware ASR and WhisperX-Anchored Pyannote Diarization for Long-Form Bengali Speech