Refactor Text2Speech: Keeping speech in memory #11
Conversation
| from pathlib import Path | ||
|
|
||
|
|
||
| def save_audio(audio: bytes) -> str: |
There was a problem hiding this comment.
It seems bizzare for me, to have this function and use namedTemporaryFile, why it cannot be user provied path?
On the other hand I generally think that likely langchain does not need to have save/load audio functionality - it is not core and maintenance burden. Can we drop it?
Also it is in wrong place - should be moved to utilities.
There was a problem hiding this comment.
For example it only works with wav files, someone serious with playing sounds should use dedicated library which can handle many different formats.
There was a problem hiding this comment.
I would left those functions in docs - to show how to do it, but would not necessary make them langchain functions, but maybe it is used by some agents so it needs to stay...
| f.write(speech) | ||
| return f.name | ||
| self.play(speech) | ||
| return "Speech has been generated" |
There was a problem hiding this comment.
I think it should just return None or just speech as bytes?
After reading code, I see it was created as a tool - it means it was crafted to be directly used within agents etc. I removed my other comments.
I think the best usage would be to have a single tool which can have different implementation provided, similar to PythonREPLTool.
Then you can have utilities which use more useful API to operate with elevenlabs/azure, just an idea.
|
|
||
| else: | ||
| return f"Speech synthesis failed: {result.reason}" | ||
| raise RuntimeError(f"Speech synthesis failed: {result.reason}") |
There was a problem hiding this comment.
Aren't you breaking agents by those changes?
add random policy and notebook example
Description:
Keeps speech in memory instead of saving it into the temporary file.
Twitter handle:
@deepsense_ai, @matt_wosinski