Update constants.py
Browse files- constants.py +30 -8
constants.py
CHANGED
|
@@ -12,14 +12,6 @@ banner_url = "https://huggingface.co/datasets/reach-vb/random-images/resolve/mai
|
|
| 12 |
BANNER = f'<div style="display: flex; justify-content: space-around;"><img src="{banner_url}" alt="Banner" style="width: 40vw; min-width: 300px; max-width: 600px;"> </div>'
|
| 13 |
|
| 14 |
EXPLANATION = """
|
| 15 |
-
## Why EdAcc Matters for ASR Evaluation
|
| 16 |
-
|
| 17 |
-
The EdAcc dataset is specifically designed to evaluate the robustness of Automatic Speech Recognition (ASR) models across diverse accents and demographics. This leaderboard helps you:
|
| 18 |
-
|
| 19 |
-
* **Assess Accent Fairness**: Compare model performance across 30+ different accents and speaker demographics
|
| 20 |
-
* **Evaluate Real-World Robustness**: Understand how ASR models perform beyond standard benchmarks
|
| 21 |
-
* **Make Informed Choices**: Select models that work well for your target demographics
|
| 22 |
-
|
| 23 |
### How to Read the Results
|
| 24 |
* **Average WER ⬇️**: Lower Word Error Rate (WER) is better
|
| 25 |
* **Average per Gender**: Average WER for each gender
|
|
@@ -29,6 +21,36 @@ EXPLANATION = """
|
|
| 29 |
Use the column filter to focus on specific demographics or view all results together.
|
| 30 |
"""
|
| 31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
TITLE = "<html> <head> <style> h1 {text-align: center;} </style> </head> <body> <h1> 🤗 Open Automatic Speech Recognition Leaderboard </b> </body> </html>"
|
| 33 |
|
| 34 |
INTRODUCTION_TEXT = "📐 Results on [EdAcc Dataset](https://huggingface.co/datasets/edinburghcstr/edacc) split by accents and gender. \
|
|
|
|
| 12 |
BANNER = f'<div style="display: flex; justify-content: space-around;"><img src="{banner_url}" alt="Banner" style="width: 40vw; min-width: 300px; max-width: 600px;"> </div>'
|
| 13 |
|
| 14 |
EXPLANATION = """
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
### How to Read the Results
|
| 16 |
* **Average WER ⬇️**: Lower Word Error Rate (WER) is better
|
| 17 |
* **Average per Gender**: Average WER for each gender
|
|
|
|
| 21 |
Use the column filter to focus on specific demographics or view all results together.
|
| 22 |
"""
|
| 23 |
|
| 24 |
+
EXPLANATION_EDACC = """
|
| 25 |
+
## EdAcc: Evaluating ASR Models Across Global English Accents
|
| 26 |
+
|
| 27 |
+
The Edinburgh International Accents of English Corpus (EdAcc) features over 40 distinct English accents from both native (L1) and non-native (L2) speakers. This evaluation helps you:
|
| 28 |
+
|
| 29 |
+
* **Compare Gender Performance**: Analyze how models perform across male and female speakers
|
| 30 |
+
* **Evaluate Regional Robustness**: Test model accuracy across European, Asian, African, and American accents
|
| 31 |
+
* **Assess Real-World Applicability**: Understand performance in natural conversational settings
|
| 32 |
+
|
| 33 |
+
The results show that:
|
| 34 |
+
* Larger models consistently outperform their smaller counterparts
|
| 35 |
+
* Multilingual models often handle accent diversity better than English-only variants
|
| 36 |
+
* Distilled models maintain good performance but show slight degradation on challenging accents
|
| 37 |
+
"""
|
| 38 |
+
|
| 39 |
+
EXPLANATION_AFRI = """
|
| 40 |
+
## AfriSpeech: Testing ASR Robustness on African English Accents
|
| 41 |
+
|
| 42 |
+
The AfriSpeech Out-of-Distribution (OOD) test set features 20 distinct African English accents not present in common training data. This benchmark:
|
| 43 |
+
|
| 44 |
+
* **Challenges Model Generalization**: Tests performance on truly underrepresented accents
|
| 45 |
+
* **Reveals Robustness Gaps**: Highlights limitations in current ASR systems
|
| 46 |
+
* **Guides Improvement**: Identifies areas needing focused development
|
| 47 |
+
|
| 48 |
+
Key findings show:
|
| 49 |
+
* Full-sized models significantly outperform distilled versions
|
| 50 |
+
* Multilingual models demonstrate better generalization to African accents
|
| 51 |
+
* Even top performers show room for improvement on these challenging accents
|
| 52 |
+
"""
|
| 53 |
+
|
| 54 |
TITLE = "<html> <head> <style> h1 {text-align: center;} </style> </head> <body> <h1> 🤗 Open Automatic Speech Recognition Leaderboard </b> </body> </html>"
|
| 55 |
|
| 56 |
INTRODUCTION_TEXT = "📐 Results on [EdAcc Dataset](https://huggingface.co/datasets/edinburghcstr/edacc) split by accents and gender. \
|