Research map / Label-efficient and robust learning / Recognition and long-tail
A Toolbox for Construction and Analysis of Speech Datasets
Evelina Bakhturina, Vitaly Lavrukhin, Boris Ginsburg
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add RuLS data to MCV-Ru for fine-tuning the Russian ASR model | QuartzNet-15x5 (Russian) · MCV Russian dev | greedy WER (%): 12.63 -> 9.65 | Improved |
| 3-fold cross-validation re-alignment with re-tuned ASR model (ablation) | CTC-Segmentation on RuLS · RuLS books | average book WER (%): 45.8 -> 18.83 | Improved |
| filter MCV English utterances by CER threshold of 25% | MCV English 7.0 validated · MCV English | duration (hours): 1 970 -> 1 938 | Mixed |
| filter MLS English train by CER threshold of 10% | MLS English train · MLS English | duration (hours): 44659 -> 42967 | Mixed |
| text normalization of MCV English (remove punctuation, drop foreign/special characters) | MCV English 7.0 validated · MCV English | alphabet size: 149 -> 28 | Improved |
| re-segment LibriSpeech dev-clean with CTC-Segmentation instead of original segmentation | LibriSpeech dev-clean · LibriSpeech dev-clean | WER (%): 3.78 -> 3.82 | No clear effect |
| re-segment LibriSpeech dev-clean with CTC-Segmentation instead of original segmentation | LibriSpeech dev-clean · LibriSpeech dev-clean | CER (%): 1.14 -> 1.16 | No clear effect |
| split sentences in the middle to shorten utterances (ablation) | RuLS text pre-processing · RuLS segmentation | segmentation quality | Hurt |
| run aligner multiple times with window sizes 8000, 10000, 12000 and keep matched segments (ablation) | NeMo CTC-Segmentation · RuLS segmentation | alignment errors | Improved |
Closest papers by meaning
- BIGOS V2 Benchmark for Polish ASR: Curated Datasets and Tools for Reproducible Evaluation (NeurIPS 2024 · similarity 0.87)
- WorldSpeech: A Multilingual Speech Corpus from Around the World (NeurIPS 2026 · similarity 0.87)
- EuroSpeech: A Multilingual Speech Corpus (NeurIPS 2025 · similarity 0.86)
- IndicVoices-R: Unlocking a Massive Multilingual Multi-speaker Speech Corpus for Scaling Indian TTS (NeurIPS 2024 · similarity 0.84)
- Speech Recognition Meets Large Language Model: Benchmarking, Models, and Exploration (AAAI 2025 · similarity 0.84)
- DataComp-LM: In search of the next generation of training sets for language models (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
