Datasets:
README.md
CHANGED
|
@@ -1,3 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# MANGO: A Corpus of Human Ratings for Speech
|
| 2 |
|
| 3 |
**MANGO** (*MUSHRA Assessment corpus using Native listeners and Guidelines to understand human Opinions at scale*) is the first large-scale dataset designed for evaluating Text-to-Speech (TTS) systems in Indian languages.
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- hi
|
| 4 |
+
- ta
|
| 5 |
+
license: cc-by-4.0
|
| 6 |
+
size_categories:
|
| 7 |
+
- 100K<n<1M
|
| 8 |
+
task_categories:
|
| 9 |
+
- text-to-speech
|
| 10 |
+
pretty_name: MANGO
|
| 11 |
+
dataset_info:
|
| 12 |
+
features:
|
| 13 |
+
- name: Rater_ID
|
| 14 |
+
dtype: int64
|
| 15 |
+
- name: FS2_Score
|
| 16 |
+
dtype: int64
|
| 17 |
+
- name: VITS_Score
|
| 18 |
+
dtype: int64
|
| 19 |
+
- name: ST2_Score
|
| 20 |
+
dtype: int64
|
| 21 |
+
- name: ANC_Score
|
| 22 |
+
dtype: int64
|
| 23 |
+
- name: REF_Score
|
| 24 |
+
dtype: int64
|
| 25 |
+
- name: FS2_Audio
|
| 26 |
+
dtype: string
|
| 27 |
+
- name: VITS_Audio
|
| 28 |
+
dtype: string
|
| 29 |
+
- name: ST2_Audio
|
| 30 |
+
dtype: string
|
| 31 |
+
- name: ANC_Audio
|
| 32 |
+
dtype: string
|
| 33 |
+
- name: REF_Audio
|
| 34 |
+
dtype: string
|
| 35 |
+
- name: ID
|
| 36 |
+
dtype: int64
|
| 37 |
+
splits:
|
| 38 |
+
- name: Tamil__MUSHRA_DG_NMR
|
| 39 |
+
num_bytes: 523400
|
| 40 |
+
num_examples: 2000
|
| 41 |
+
- name: Hindi__MUSHRA_DG
|
| 42 |
+
num_bytes: 563800
|
| 43 |
+
num_examples: 2000
|
| 44 |
+
- name: Hindi__MUSHRA_NMR
|
| 45 |
+
num_bytes: 2875380
|
| 46 |
+
num_examples: 10200
|
| 47 |
+
- name: Hindi__MUSHRA_DG_NMR
|
| 48 |
+
num_bytes: 563800
|
| 49 |
+
num_examples: 2000
|
| 50 |
+
- name: Tamil__MUSHRA_NMR
|
| 51 |
+
num_bytes: 2538490
|
| 52 |
+
num_examples: 9700
|
| 53 |
+
- name: Tamil__MUSHRA_DG
|
| 54 |
+
num_bytes: 523400
|
| 55 |
+
num_examples: 2000
|
| 56 |
+
- name: Tamil__MUSHRA
|
| 57 |
+
num_bytes: 2617000
|
| 58 |
+
num_examples: 10000
|
| 59 |
+
- name: Hindi__MUSHRA
|
| 60 |
+
num_bytes: 3185470
|
| 61 |
+
num_examples: 11300
|
| 62 |
+
download_size: 1045866
|
| 63 |
+
dataset_size: 13390740
|
| 64 |
+
configs:
|
| 65 |
+
- config_name: default
|
| 66 |
+
data_files:
|
| 67 |
+
- split: Tamil__MUSHRA_DG_NMR
|
| 68 |
+
path: csvs/tamil_mushra_dg_nmr*
|
| 69 |
+
- split: Hindi__MUSHRA_DG
|
| 70 |
+
path: csvs/hindi_mushra_dg*
|
| 71 |
+
- split: Hindi__MUSHRA_NMR
|
| 72 |
+
path: csvs/hindi_mushra_nmr*
|
| 73 |
+
- split: Hindi__MUSHRA_DG_NMR
|
| 74 |
+
path: csvs/hindi_mushra_dg_nmr*
|
| 75 |
+
- split: Tamil__MUSHRA_NMR
|
| 76 |
+
path: csvs/tamil_mushra_nmr*
|
| 77 |
+
- split: Tamil__MUSHRA_DG
|
| 78 |
+
path: csvs/tamil_mushra_dg*
|
| 79 |
+
- split: Tamil__MUSHRA
|
| 80 |
+
path: csvs/tamil_mushra*
|
| 81 |
+
- split: Hindi__MUSHRA
|
| 82 |
+
path: csvs/hindi_mushra*
|
| 83 |
+
tags:
|
| 84 |
+
- speech
|
| 85 |
+
- evaluation
|
| 86 |
+
- mushra
|
| 87 |
+
---
|
| 88 |
+
|
| 89 |
# MANGO: A Corpus of Human Ratings for Speech
|
| 90 |
|
| 91 |
**MANGO** (*MUSHRA Assessment corpus using Native listeners and Guidelines to understand human Opinions at scale*) is the first large-scale dataset designed for evaluating Text-to-Speech (TTS) systems in Indian languages.
|