Update README.md

README.md

@@ -6,14 +6,18 @@ task_categories:
 
 🏆 **News:** Our [OWSM v4 paper](https://www.isca-archive.org/interspeech_2025/peng25c_interspeech.html) won the [Best Student Paper Award](https://isca-speech.org/ISCA-Awards) at INTERSPEECH 2025!
 
-This dataset is the new training data of [OWSM v4 models](https://www.isca-archive.org/interspeech_2025/peng25c_interspeech.html). 
-It is a high-quality subset of [YODAS2](https://huggingface.co/datasets/espnet/yodas2), comprising 166,000 hours of multilingual speech spanning 75 languages.
+[Open Whisper-style Speech Model (OWSM)](https://www.wavlab.org/activities/2024/owsm/) is the first **fully open** Whisper-style speech foundation model. 
+It reproduces and advances OpenAI's Whisper-style training using publicly available data and open-source toolkits. 
+The code, pre-trained model weights, and training logs are publicly released to promote open science in speech foundation models.
+
+This repo contains the newly curated training data for [OWSM v4](https://www.isca-archive.org/interspeech_2025/peng25c_interspeech.html), the latest version in the OWSM series.
+This dataset is a high-quality subset of [YODAS2](https://huggingface.co/datasets/espnet/yodas2), comprising 166,000 hours of multilingual speech spanning 75 languages.
 Utterances are segmented into up to 30 seconds to be consistent with OWSM training.
 
 ## YODAS Data Cleaning
 
 Due to the nature of web-sourced data, the original YODAS2 dataset contains inaccurate language labels and misaligned audio-text pairs. 
-Our preliminary experiments suggest that such noise has a negative impact on the performance of downstream ASR models. 
+Our preliminary experiments suggest that such noise hurts the performance of downstream ASR models. 
 To address this, we developed a scalable data-cleaning pipeline using publicly available toolkits, resulting in a curated subset of the original dataset. 
 This cleaned dataset forms a core part of the training data for our OWSM v4 models, which, when combined with existing OWSM data, significantly outperform previous versions on multilingual benchmarks.
 
@@ -21,14 +25,40 @@ This cleaned dataset forms a core part of the training data for our OWSM v4 mode
 - **Data Cleaning Scripts:** [ESPnet](https://github.com/espnet/espnet/tree/master/egs2/owsm_v4/s2t1)
 - **Model Demo:** [Gradio](https://huggingface.co/spaces/espnet/OWSM_V4_Demo)
 
+The data cleaning process consists of three stages.
 
-## OWSM v4 Results
+![YODAS Data Cleaning Procedure](data-cleaning-stats.png)
 
-[Open Whisper-style Speech Model (OWSM)](https://www.wavlab.org/activities/2024/owsm/) is the first **fully open** Whisper-style speech foundation model. 
-It reproduces and advances OpenAI's Whisper-style training using publicly available data and open-source toolkits. 
-The code, pre-trained model weights, and training logs are publicly released to promote open science in speech foundation models.
+### Stage 1: Resegmentation (Section 2.1.1 in Paper)
+
+YODAS provides unsegmented long-form recordings, each of which is accompanied by a list of text transcriptions annotated with start and end timestamps. 
+However, some timestamps are inaccurate. Consequently, our first step is to realign the audio and text using the CTC segmentation algorithm.
+
+### Stage 2: LID-based filtering (Section 2.1.2 in Paper)
+
+Some utterances have incorrect language labels. We perform language identification on both audio and text using public models. 
+Then, we remove utterances where the language label does not match the identified language from either audio or text.
+
+### Stage 3: CTC-score-based filtering (Section 2.1.3 in Paper)
+
+The CTC segmentation algorithm in Stage 1 assigns a confidence score to each utterance, which measures the speech-text alignment quality. 
+We filter out utterances with low CTC scores. 
+The CTC confidence score is language-dependent; therefore, we rank the scores of short utterances within each language and select a relative threshold (quantile).
+
+In this repo, we release two subsets:
+- `dump/raw/yodas0.00` (low quality, not recommended): The filtering threshold is 0.00, i.e., no filtering based on CTC score. This subset contains all data at the end of Stage 2.
+  - ⚠️ **This subset has low-quality data that hurts ASR performance, as shown in Table 2 of our paper. It is NOT recommended to use unless further filtering is performed.**
+- `dump/raw/yodas0.10` (good quality, recommended): The filtering threshold is 0.10. This is the actual training data used to develop OWSM v4.
+
+## Usage
+
+This dataset follows the ESPnet OWSM data format as described in the [`s2t1` recipe](https://github.com/espnet/espnet/tree/master/egs2/TEMPLATE/s2t1).
+Two subsets are released:
+
+
+## OWSM v4 Results
 
-[OWSM v4](https://www.isca-archive.org/interspeech_2025/peng25c_interspeech.html) is the latest version in the OWSM series, which significantly outperforms prior versions.
+[OWSM v4](https://www.isca-archive.org/interspeech_2025/peng25c_interspeech.html) is 
 
 Please refer to our paper for comprehensive evaluations. Below are some notable results.