-
Notifications
You must be signed in to change notification settings - Fork 0
September 28th Meeting
Stefan Scodellaro edited this page Sep 29, 2023
·
6 revisions
- Research
- Clean up and add to each others sections
- Proposal (September 16/17)
- Project Requirements
- Conclusion
- Implementation (September 28)
- Speech To Text API Research
- Proposal (September 16/17)
- State of Art
- Project Management Plan
- Implementation (September 28)
- LLM API Research
- Proposal (September 16/17)
- Problem Statement
- Scope
- Implementation (September 28)
- Text to Speech API Research
Research Notes:
- One interesting thing is "streaming mode" -> API returns text as soon as text is generated instead of waiting for the whole prompt.
- OpenAI always takes atleast one second.
Demo of google cloud pipeline:
- Complete speech to text, LLM generation and text to audio output.
- Entire pipeline using one API.
- Google PaLM2 model isn't as good as GPT-3.5. For example, inputting "hello" with default settings generates a hello world C++ program.
Marc began initial investigation for android development. The entire team will continue learning Kotlin (the language android apps are programmed in) for the next meeting.
- By doing reflections, we've expanded into a component of all talk therapy
- Need to focus on story since we are already good at technical aspects
- In future meetings, decide who says what in advance
- Speech diarization research will come later
- Need more research for text to speech inflections
- Decide if "end of input" determination is a problem
- "What" we do is more important than reports
- Need to figure out meeting 1 details
- Rose recommends reading this article