Update Dataset Card: release 1.1.0

README.md

@@ -2,7 +2,7 @@
 language:
 - bm
 pretty_name: Transcription Scorer
-version: 1.0.0
+version: 1.1.0
 tags:
 - audio
 - speech
@@ -11,7 +11,7 @@ tags:
 - ASR
 - reward-model
 - Bambara
-license: cc-by-4.0
+license: cc-by-sa-4.0
 task_categories:
 - automatic-speech-recognition
 - reinforcement-learning
@@ -30,14 +30,12 @@ dataset_info:
     dtype: audio
   - name: duration
     dtype: float
-  - name: transcription
+  - name: text
     dtype: string
   - name: score
     dtype: float
-  - name: labeler
-    dtype: string
   total_audio_files: 2153
-  total_duration_hours: ~9
+  total_duration_hours: ~2
 - config_name: partially-reviewed
   features:
   - name: audio
@@ -50,13 +48,13 @@ dataset_info:
     dtype: float64
   splits:
   - name: train
-    num_bytes: 600583588.0
+    num_bytes: 600583588
     num_examples: 1000
   - name: test
-    num_bytes: 116626924.0
+    num_bytes: 116626924
     num_examples: 200
   download_size: 695513651
-  dataset_size: 717210512.0
+  dataset_size: 717210512
 configs:
 - config_name: partially-reviewed
   data_files:
@@ -68,36 +66,38 @@ configs:
 
 # Transcription Scorer Dataset
 
-The **Transcription Scorer** dataset was created to support research in reference-free evaluation of Automatic Speech Recognition (ASR) systems using **human feedback**. Unlike traditional evaluation metrics such as WER and its derivatives, this dataset reflects subjective judgments of ASR outputs by human raters across multiple criteria, simulating the way a teacher grades students.
+The **Transcription Scorer** dataset was created to support research in reference-free evaluation of Automatic Speech Recognition (ASR) systems using **human feedback**. Unlike traditional evaluation metrics such as WER and its derivatives, this dataset reflects judgments of ASR outputs by human raters across multiple criteria, simulating the way a teacher grades students.
 
 ## ⚙️ What’s Inside
 
-This dataset contains **2153 audio samples** (from diverse sources including speech and music with lyrics), each associated with:
+This dataset contains **1200 audio samples** (from diverse sources including music with lyrics) totaling 2.28 hours. It is made of short to meduim length segments each associated with:
 
-- One **transcriptions** (By selecting the best hypothesis of different models)
+- One **transcriptions** (drawn by selecting the best hypothesis of two Bambara ASR models)
 - A **score** between 0 and 100 assigned by human annotators
-- The **labeler** identity (What generated the transcription. That wasn't included during scoring)
+
+| bucket (s) | partially‑reviewed |
+| ---------- | --------------     |
+| 0.6 – 15   | 965                |
+| 15 – 30    | 235                |
 
 ### Sources:
 - Transcriptions were generated by two ASR models: 
-  - **Djelia-V1** (proprietary, API-based)
+  - **Djelia-V1** (proprietary, access through API)
   - **Soloni** (open-source from [RobotsMali](https://huggingface.co/RobotsMali/soloni-114m-tdt-ctc-V0))
-- Additional 151 transcriptions were human-generated
-- 100 transcriptions were intentionally **randomized/shuffled** to measure baseline judgment.
+- Additional 81 transcriptions were intentionally **randomized/shuffled** to measure baseline judgment.
 
-Most of the audios were collected by RobotsMali AI4D Lab with the [Office de Radio et Télévision du Mali](https://www.ortm.ml/) which gave us early access to a few archives of some of their past emissions in Bamanankan. But this dataset also include a few samples from [Jeli-ASR](https://huggingface.co/datasets/RobotsMali/jeli-asr) and some in house data that had not been published up to now (the 100 human made transcriptions)
+Most of the audios were collected by RobotsMali AI4D Lab with the [Office de Radio et Télévision du Mali](https://www.ortm.ml/) which gave us early access to a few archives of some of their past emissions in Bamanankan. But this dataset also include a few samples from [bam-asr-early](https://huggingface.co/datasets/RobotsMali/bam-asr-early). 
 
-The evaluation was based on this criteria document: [Scoring Guide](https://docs.google.com/document/d/e/2PACX-1vRHFEAwU4C43NUHEY85auokgiG9dJgB0ApKwY41fwFGYn7xUSl1hXnk-CBp0_67c1C7mC7jXLzte3Mu/pub), but annotators had freedom to assign scores as they would in a real-world grading scenario. So it is a Human feedback dataset but not based on preferences only, the score is actually designed to be a valuable quality metric. 
+The evaluation was based on the [following criteria](https://docs.google.com/document/d/e/2PACX-1vRHFEAwU4C43NUHEY85auokgiG9dJgB0ApKwY41fwFGYn7xUSl1hXnk-CBp0_67c1C7mC7jXLzte3Mu/pub) but we also left room for a personnal subjective judgement so it also include some form of human preference feedback as the annotations were partially reviwed by professional Bambara linguists. So it is a Human feedback dataset but not based on preferences only, the score is actually designed to be a refective of the quality of the transcriptions enough to be used as a proxy metric.
 
 ## **Usage**
 
-This dataset is intended for researchers and developers who face a label scarcity situation making traditional ASR evaluation metrics like WER impossible (which is especially relevent to low resource languges such as Bambara). By leveraging human-assigned scores, it enables the development of models which outputs can be used as a proxy to transcription quality. Whether you're building evaluation tools or studying human feedback in speech systems, this dataset provides a rich ground for experimentation.
+This dataset is intended for researchers and developers who face a label scarcity situation making traditional ASR evaluation metrics like WER impossible (which is especially relevent to low resource languges such as Bambara). By leveraging human-assigned scores, it enables the development of scoring models which outputs can be used as a proxy to transcription quality. Whether you're building evaluation tools or studying human feedback in speech systems, you might find this dataset useful if you face label scarcity.
 
 - Developing **reference-free** evaluation metrics  
-- Training **reward models** for RLHF-based fine-tuning of ASR systems  
+- Training **reward models** for RLHF-based fine-tuning of ASR systems
 - Understanding how **human preferences** relate to transcription quality  
 
-The data is in .arrow format for compatibility with HF's Datasets Library. So you don't need any ajustement to load the dataset directly with datasets:
 
 ```python
 from datasets import load_dataset
@@ -109,10 +109,10 @@ dataset = load_dataset("RobotsMali/transcription-scorer", "partially-reviewed")
 
 ## Data Splits
 
-- **Train**: 1937 examples (~8.5h)
-- **Test**: 216 examples (~0.6h)
+- **Train**: 1000 examples (~1.92h)
+- **Test**: 200 examples (~0.37h)
 
-This initial version is only **partially reviewed**, hence the config name. We're actively seeking collaborators to help review the scores.
+This initial version is only **partially reviewed**, so you may contribute by opening a PR or a discussion if you find that some assigned scores are innacurate.
 
 ## Fields
 
@@ -120,17 +120,16 @@ This initial version is only **partially reviewed**, hence the config name. We'r
 - `duration`: audio length (seconds)
 - `transcription`: text output to be scored
 - `score`: human-assigned score (0–100)
-- `labeler`: identifier for annotator
 
-## Known Limitations
+## Known Limitations / Issues 
 
-- Human scoring may contain inconsistencies (quite natural in human eval).
+- Human scoring may contain inconsistencies.
 - Only partial review/consensus exists — **scores may be refined** in future versions.
-- Not all samples were evaluated by all annotators (Actually every sample had a unique annotator. That might change in future versions).
+- The dataset is very limited in context diversity and transcription variance, only two models were used to generate transcriptions for the same ~560 audios + 80 shuffled transcriptions for baseline estimation so it will benefit from additional data from different distribution.
 
 ## 🤝 Contribute
 
-Feel something was misjudged? Want to improve score consistency? Please open a discussion  — we **welcome feedback and collaboration**.
+Feel something was misjudged? Want to improve score consistency? Add transcriptions from another model ? Please open a discussion  — we **welcome feedback and collaboration**.
 
 ---
 
@@ -139,10 +138,10 @@ Feel something was misjudged? Want to improve score consistency? Please open a d
 ```bibtex
 @misc{transcription_scorer_2025,
   title={A Dataset of human evaluations of Automatic Speech Recognition for low Resource Bambara language},
-  author={Yacouba Diarra and Panga Azazia Kamaté and Adam Bouno Kampo and Nouhoum Coulibaly},
+  author={RobotsMali AI4D Lab},
   year={2025},
   publisher={Hugging Face}
 }
 ```
 
----
+---