LOBE is a recording client made specifically for TTS data collections. It supports multiple collections, single and multi-speaker, and can prompt sentences based on phonetic coverage.
We would like to be able to capture conversational data via LOBE. This would require at least:
A new dataset db model specifically for conversational datasets
A new / or extended recording view for conversational collection. We should display at least feedback from mic and duration of recording in this view.
The biggest concern here are:
Data integrity: we want to make sure that if an exception or any errors pop up that we loose only a limited amount of data. This requires some type of check-pointing or streaming to server.
Memory overhead on client: Using some features in LOBE for an extended period of time might cause some memory overhead on client. We should investigate possible overheads and address them.
We would like to be able to capture conversational data via LOBE. This would require at least:
A new dataset db model specifically for conversational datasets A new / or extended recording view for conversational collection. We should display at least feedback from mic and duration of recording in this view.
The biggest concern here are: