MultIPA is yet another automatic speech transcription model into phonetic IPA. The idea is that, if we train a multilingual speech-to-IPA model with enough amount of good phoneme representations, the model's output will be approximated to phonetic transcriptions.
Note that the codes in this repository are incomplete; we are still cleaning up the original codes for publishing in this repository. The finalized codes will be prepared by the day of the presentation at INTERSPEECH 2023 (mid-August). We appreciate your patience!
At this moment, we have the following languages incorporated available in the training data:
- English
- Finnish
- Hungarian
- Japanese
- Maltese
- Modern Greek
- Polish
- Tamil
Note that English was added after the INTERSPEECH 2023 paper was submitted. We aim to include more languages to take into account linguistic diversity.