Speech-to-IPA Conversion
- aramhayr
- Feb 9
- 1 min read
Speech-to-IPA conversion seems very useful as a research tool specifically for languages that have no writing system. Graphical (in IPA) recording of these languages will allow statistical and NLP methods to be applied to them: tree-banks can be created.
It is worth investigating if it makes sense for currently used speech-to-text system adding an extra conversion layer: Speech-to-IPA-<language recognition>-to-scriptBasedText.
In fact the two-layer architecture of current implementation can be preserved if the acoustic model outputs phonetic information in IPA, while the language model receives the input in IPA.
Comments