Update README

README.md

@@ -1,5 +1,5 @@
 ---
-license: mit
+license: cc-by-4.0
 configs:
 - config_name: default
   data_files:
@@ -35,25 +35,33 @@ dataset_info:
     num_examples: 45
   download_size: 19975985
   dataset_size: 20690456
+task_categories:
+- automatic-speech-recognition
+language:
+- bm
+tags:
+- speech
+- asr
+- bambara
+- low-resource
 ---
 # Nyana-Eval Dataset
 
 ## Dataset Description
 
-**Nyana-Eval** is a compact, stratified evaluation subset for benchmarking Automatic Speech Recognition (ASR) models in Bambara. It consists of **45 audio recordings** totaling approximately **3.03 minutes** (0.05 hours), carefully selected to represent real-world linguistic and acoustic challenges in low-resource Bambara speech. This dataset is derived from the larger [RobotsMali/Bam_ASR_Eval_500](https://huggingface.co/datasets/RobotsMali/Bam_ASR_Eval_500) corpus and is optimized for quick, reproducible ASR testing.
+**Nyana-Eval** is a compact, stratified evaluation subset for benchmarking Automatic Speech Recognition (ASR) models in Bambara. It consists of **45 audio recordings** totaling approximately **3.03 minutes**, carefully selected to represent real-world linguistic and acoustic challenges in low-resource Bambara speech. This dataset is derived from the larger [RobotsMali/Bam_ASR_Eval_500](https://huggingface.co/datasets/RobotsMali/Bam_ASR_Eval_500) corpus and is optimized for quick, reproducible human evaluation.
 
 Nyana-Eval is ideal for:
 - Rapid evaluation of Bambara ASR models (e.g., WER/CER computation on diverse conditions).
 - Human-assisted qualitative analysis (e.g., semantic fidelity, code-switching handling).
-- Addressing gaps in low-resource settings: dialectal variations, noise, proper names, and code-mixing with French.
+- Testing models on low-resource settings gaps: dialectal variations, noise, proper names, and code-mixing with French.
 
 **Key Statistics**:
 - **Total Samples**: 45 (balanced: 15 per source subset).
 - **Total Duration**: ~3.03 minutes (average ~4.0 seconds per sample).
-- **Audio Format**: Mono-channel WAV files at 16 kHz sampling rate.
+- **Audio Format**: Mono-channel WAV files at 16 | 44.1k kHz sampling rate.
 - **Languages**: Primary: Bambara (Bamana); Secondary: French code-switching (~15% of samples).
-- **License**: MIT License (open for research, commercial use with attribution).
-- **Tags**: `audio`, `speech`, `asr`, `bambara`, `bamanan`, `low-resource`, `evaluation`, `human-evaluation`, `african-languages`.
+- **License**: CC-BY-4.0 License (open for research, commercial use with attribution).
 
 Compiled by Robots Mali AI4D Lab, this dataset powers the human-comparative analysis in the [Bambara ASR Models Evaluation Report].
 
@@ -65,10 +73,10 @@ Nyana-Eval is a single-split dataset (default: `test`), with each entry includin
 
 | Column          | Type     | Description                                                                 | Example Value |
 |-----------------|----------|-----------------------------------------------------------------------------|---------------|
-| `audio`        | Audio   | Raw audio waveform (array + sampling rate: 16 kHz) or file path.            | `{"path": "1.1.wav", "array": [...], "sampling_rate": 16000}` |
-| `duration`     | Float64 | Length of the audio clip in seconds (range: 0.62s – 15s).                   | 3.45         |
-| `references`   | String  | Bambara text                                                                | "nɔgɔ ye a ka tɔɔrɔ ye" |
-| '8 * models transcriptions' | String  | Bambara text  transcription                                    |  |
+| `audio`        | Audio   | Raw audio waveform (array + sampling rate: 16 | 44.1k kHz) or file path.      | `{"path": "1.1.wav", "array": [...], "sampling_rate": 16000}` |
+| `duration`     | Float64 | Length of the audio clip in seconds (range: 0.62s – 15s).                     | 3.45         |
+| `references`   | String  | Bambara text                                                                  | "nɔgɔ ye a ka tɔɔrɔ ye" |
+| '8 * models transcriptions' | String  | ASR provised transcriptions                                      |  |
 
 ### Splits
 - **Default Split**: Full 45 samples (`test` for evaluation).
@@ -84,17 +92,17 @@ print(dataset[0])    # Example: {'audio': ..., 'duration': 3.45, 'transcription'
 
 ## Sources and Compilation
 
-Nyana-Eval is a **balanced subsample (15 per subset)** from the full 500-sample [RobotsMali/Bam_ASR_Eval_500](https://huggingface.co/datasets/RobotsMali/Bam_ASR_Eval_500) corpus (~36.69 minutes total). Selection criteria ensured diversity: voice variety (age/gender/accents), acoustic challenges (noise/volume/overlaps), and linguistic phenomena (code-switching, proper names, proverbs).
+Nyana-Eval is a **balanced subsample (15 per subset)** from the full 500-sample [RobotsMali/Bam_ASR_Eval_500](https://huggingface.co/datasets/RobotsMali/Bam_ASR_Eval_500) corpus (~36.69 minutes total). Selection criteria ensured diversity: voice variety (age/gender/accents), acoustic challenges (noise/volume/overlaps), and linguistic phenomena (code-switching, proper names etc.)
 
 **Parent Subsets Breakdown** (15 samples each in Nyana-Eval):
 - **Ref. 1: RobotsMali/kunkado (Hugging Face)** – 15 audios (~1.96 minutes scaled).  
   Semi-supervised interviews and spontaneous discourse. Source: [RobotsMali/kunkado](https://huggingface.co/datasets/RobotsMali/kunkado). Focus: Dialectal variations and natural flow.
 
-- **Ref. 2: jeli-ASR street interviews subset** – 15 audios (~0.31 minutes scaled).  
-  Griot traditions and urban interviews. Emphasizes oral storytelling, cultural terms, and hesitations.
+- **Ref. 2: jeli-ASR street interviews subset** – 30 audios (~1.85 minutes).  
+  Street interviews Subset from the jeli-asr project. Source: [jeli-asr](https://github.com/robotsmali-ai/jeli-asr/)
 
-- **Ref. 3: Extracts from An Bɛ Kalan app (Robots Mali)** – 15 audios (~3.34 minutes scaled).  
-  User-generated readings from the Bambara learning app. Captures learner speech with occasional errors or pauses.
+- **Ref. 3: Readings of Excerpts from An Bɛ Kalan app (RobotsMali)** – 220 audios (~20.06 minutes).  
+  User-generated readings and interactions from the mobile app for Bambara learning, captures learner speech with occasional errors or pauses. source: [https://github.com/Robotsmali-ai/an-be-kalan]
 
 ## Metadata
 
@@ -106,13 +114,10 @@ Nyana-Eval is a **balanced subsample (15 per subset)** from the full 500-sample
 - **Download Size**: ~25 MB (audios + metadata).
 - **Ethical Notes**: Ethically sourced/anonymized; focuses on public-domain cultural speech. For research; cite Robots Mali.
 
-### Linguistic Metadata
-- **Dialect Coverage**: Urban Bamana (Bamako-influenced) with rural elements.
-- **Phonetic Coverage**: Tones, nasals, contractions; with OOD proper names (e.g., "Sunjata," "Traoré").
-- **Challenges Represented**:
+### Challenges Represented:
   - Code-switching: samples (e.g., "Segou ville").
+  - Proper names (e.g., "Sunjata," "Traoré").
   - Noise/Overlaps: (e.g., low-volume interviews, multi-speaker).
-  - Human Eval Focus: 45 samples scored 0-3 + bonuses for fidelity, names, switching, robustness.
 
 ### Evaluation Metadata
 - **WER Baselines**: From report – Soloni-v2: 36.07%; QuartzNet-v0: 65.42% (greedy decoding).
@@ -127,6 +132,6 @@ Nyana-Eval is a **balanced subsample (15 per subset)** from the full 500-sample
 ## Related Resources
 - **Parent Dataset**: [RobotsMali/Bam_ASR_Eval_500](https://huggingface.co/datasets/RobotsMali/Bam_ASR_Eval_500) (full 500 samples).
 - **Models**: Test with [RobotsMali ASR models](https://huggingface.co/RobotsMali/models)
-- **App**: Collect similar data via [An Bɛ Kalan](https://play.google.com/store/apps/details?id=com.robotsmali.anbekalan).
+- **App**: Collect similar data via [An Bɛ Kalan](https://play.google.com/store/apps/details?id=org.robotsmali.literacy_app).
 
-This README is self-contained; explore the attached report PDF for detailed human annotations and model rankings on these exact 45 samples!
+This README is self-contained; explore the attached report PDF for detailed human annotations and model rankings on these exact 45 samples!