TTSDS Dataset legend
Browse files
README.md
CHANGED
@@ -106,6 +106,35 @@ Example if the proprietary ElevenLabs were to be added to the capabilities table
|
|
106 |
|
107 |
More info on how the capabilities table came about can be found within the [GitHub Issue](https://github.com/Vaibhavs10/open-tts-tracker/issues/14).
|
108 |
|
109 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
110 |
|
111 |
---
|
|
|
106 |
|
107 |
More info on how the capabilities table came about can be found within the [GitHub Issue](https://github.com/Vaibhavs10/open-tts-tracker/issues/14).
|
108 |
|
109 |
+
## train_data Legend
|
110 |
+
|
111 |
+
Legend for the separate TTSDS Datasets ([_train_data_ viewer](https://huggingface.co/datasets/Pendrokar/open_tts_tracker/viewer/train_data) [GitHub](https://github.com/ttsds/ttsds_systems))
|
112 |
+
|
113 |
+
- π Multilingual
|
114 |
+
- The ISO codes of languages the model is capable off. β if English only.
|
115 |
+
- π Training Amount (k hours)
|
116 |
+
- The number of hours the model was trained on
|
117 |
+
- π§ Num. Parameters (M)
|
118 |
+
- How many parameters the model has, excluding vocoder and text-only components
|
119 |
+
- π― Target Repr.
|
120 |
+
- Which output representations the model uses, for example audio codecs or mel spectrograms
|
121 |
+
- π LibriVox Only
|
122 |
+
- If the model was trained on librivox-like (audiobook) data alone
|
123 |
+
- π NAR
|
124 |
+
- If the model has a significant non-autoregressive component
|
125 |
+
- π AR
|
126 |
+
- If the model has a significant autoregressive component
|
127 |
+
- π‘ G2P
|
128 |
+
- If the model uses G2P (phone inputs)
|
129 |
+
- 𧩠Language Model
|
130 |
+
- If an LM-like approach is used (next token prediction)
|
131 |
+
- π΅ Prosody Prediction
|
132 |
+
- If prosodic correlates such as pitch or energy are predicted
|
133 |
+
- π Diffusion
|
134 |
+
- If diffusion is used (outside vocoder)
|
135 |
+
- β±οΈ Delay Pattern
|
136 |
+
- If a delay pattern is used for audio codes (see [Lyth & King, 2024](https://arxiv.org/abs/2402.01912))
|
137 |
+
|
138 |
+
_Please create pull requests to update the info on the models!_
|
139 |
|
140 |
---
|