Describe the feature you'd like
As it stands, the API accepts a speed value between 0.25 and 4. However, in practice at speeds above 1.5 the output becomes unintelligible, and therefore not usable. What I propose is an additional optional stretching/squishing parameter to the API that allows setting of a post processing speed control. I.e. you could process the prompt at 1.0 speed, then stream the output afterwards at 2 times that speed for more natural sounding fast speech.
Describe alternatives you've considered
I have considered building a second application layer/attempting to control this in my endpoint, however, this would be hacky and less universal of a solution to a presumably common issue. Since the FastAPI wrapper is already working with raw audio before sending the result, it should be faster to complete the process earlier than later.
Describe the feature you'd like
As it stands, the API accepts a speed value between 0.25 and 4. However, in practice at speeds above 1.5 the output becomes unintelligible, and therefore not usable. What I propose is an additional optional stretching/squishing parameter to the API that allows setting of a post processing speed control. I.e. you could process the prompt at 1.0 speed, then stream the output afterwards at 2 times that speed for more natural sounding fast speech.
Describe alternatives you've considered
I have considered building a second application layer/attempting to control this in my endpoint, however, this would be hacky and less universal of a solution to a presumably common issue. Since the FastAPI wrapper is already working with raw audio before sending the result, it should be faster to complete the process earlier than later.