Update README.md
Browse files
README.md
CHANGED
|
@@ -23,6 +23,9 @@ base_model:
|
|
| 23 |
ByT5-small fine-tuned for **Grapheme-to-Phoneme (G2P)** conversion in Brazilian Portuguese, producing transcriptions in the **International Phonetic Alphabet (IPA)**.
|
| 24 |
The model is conditioned on **dialect tags** and was trained on the **BIPA** dataset (derived from Wiktionary, CC BY-SA 4.0 license), covering six dialectal varieties of Brazilian Portuguese.
|
| 25 |
|
|
|
|
|
|
|
|
|
|
| 26 |
## Quick Summary
|
| 27 |
|
| 28 |
- **Input**: short Brazilian Portuguese text, with or without a dialect tag.
|
|
@@ -97,6 +100,19 @@ Tags used in the input (from the `tag_dialeto` column of the processed dataset):
|
|
| 97 |
|
| 98 |
If no tag is provided, behavior is not guaranteed; it is always recommended to prefix the input with an explicit tag.
|
| 99 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
## Training Data
|
| 101 |
- **Source**: BIPA (Brazilian IPA), built from **Wiktionary** entries with multiple phonetic transcriptions per word and dialect.
|
| 102 |
|
|
@@ -105,9 +121,13 @@ If no tag is provided, behavior is not guaranteed; it is always recommended to p
|
|
| 105 |
- Trained exclusively on **Brazilian Portuguese**; not suitable for other languages.
|
| 106 |
- Transcriptions follow BIPA/Wiktionary-specific conventions and may differ from other phonetic standards.
|
| 107 |
- **Minority dialects**: `[regiao-nordeste]` and `[regiao-centro-oeste]` are underrepresented in the BIPA dataset compared to `[brasil]`, `[são-paulo]`, and `[rio-de-janeiro]`. Expect lower transcription accuracy for these varieties.
|
|
|
|
| 108 |
## How to Cite
|
| 109 |
|
| 110 |
-
|
|
|
|
|
|
|
|
|
|
| 111 |
|
| 112 |
- Model: `thiagomonteles/BIPA_g2p_Multidialect_Byt5`
|
| 113 |
-
- Dataset: `thiagomonteles/BIPA`
|
|
|
|
| 23 |
ByT5-small fine-tuned for **Grapheme-to-Phoneme (G2P)** conversion in Brazilian Portuguese, producing transcriptions in the **International Phonetic Alphabet (IPA)**.
|
| 24 |
The model is conditioned on **dialect tags** and was trained on the **BIPA** dataset (derived from Wiktionary, CC BY-SA 4.0 license), covering six dialectal varieties of Brazilian Portuguese.
|
| 25 |
|
| 26 |
+
paper:
|
| 27 |
+
[BIPA: Brazilian Portuguese Phonetic Dataset with Dialectal Variations in IPA Standard](https://aclanthology.org/2026.propor-1.47/)
|
| 28 |
+
|
| 29 |
## Quick Summary
|
| 30 |
|
| 31 |
- **Input**: short Brazilian Portuguese text, with or without a dialect tag.
|
|
|
|
| 100 |
|
| 101 |
If no tag is provided, behavior is not guaranteed; it is always recommended to prefix the input with an explicit tag.
|
| 102 |
|
| 103 |
+
## Performance by Dialect
|
| 104 |
+
|
| 105 |
+
Model performance (Minimum PER) on the validation set, segmented by dialect. Results show a strong correlation between training data volume and transcription accuracy — minority dialects with fewer examples present significantly higher error rates.
|
| 106 |
+
|
| 107 |
+
| Dialect | No. Examples | ↓ Min. PER (%) |
|
| 108 |
+
|---|---|---|
|
| 109 |
+
| São Paulo | 1,141 | 1.93 |
|
| 110 |
+
| Rio de Janeiro | 2,373 | 3.08 |
|
| 111 |
+
| Brazil (general) | 5,316 | 6.07 |
|
| 112 |
+
| South Region | 2,131 | 2.36 |
|
| 113 |
+
| Northeast Region | 32 | 78.12 |
|
| 114 |
+
| Center-West Region | 7 | 42.86 |
|
| 115 |
+
|
| 116 |
## Training Data
|
| 117 |
- **Source**: BIPA (Brazilian IPA), built from **Wiktionary** entries with multiple phonetic transcriptions per word and dialect.
|
| 118 |
|
|
|
|
| 121 |
- Trained exclusively on **Brazilian Portuguese**; not suitable for other languages.
|
| 122 |
- Transcriptions follow BIPA/Wiktionary-specific conventions and may differ from other phonetic standards.
|
| 123 |
- **Minority dialects**: `[regiao-nordeste]` and `[regiao-centro-oeste]` are underrepresented in the BIPA dataset compared to `[brasil]`, `[são-paulo]`, and `[rio-de-janeiro]`. Expect lower transcription accuracy for these varieties.
|
| 124 |
+
|
| 125 |
## How to Cite
|
| 126 |
|
| 127 |
+
If you use this model or the BIPA dataset, please cite the following paper:
|
| 128 |
+
|
| 129 |
+
> Thiago Monteles de Sousa, Lucas Rafael Gris, and Nádia Félix Felipe da Silva. 2026. [BIPA: Brazilian Portuguese Phonetic Dataset with Dialectal Variations in IPA Standard](https://aclanthology.org/2026.propor-1.47/). In *Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1*, pages 478–487, Salvador, Brazil. Association for Computational Linguistics.
|
| 130 |
+
|
| 131 |
|
| 132 |
- Model: `thiagomonteles/BIPA_g2p_Multidialect_Byt5`
|
| 133 |
+
- Dataset: `thiagomonteles/BIPA`
|