Q: KokoroWavSynthesizer - is it possible to know the boundaries of generated speech?

[lostmsu]

I am building a system where user can interrupt TTS. The system will need to know what was and what was not spoken. Right now I am using KokoroWavSynthesizer.Synthesize and its OnProgress callback. It, however, does not provide any information besides the generated audio. Does Kokoro provide this information at a lower level? Can KokoroSharp expose it?

[Lyrcaxis]

Hey. Yes, currently they're exposed on SynthesisHandle, which is returned by either Speak or SpeakFast.
The guess is made from SpeechGuesser. It's a rough estimate right now (only the LowEffort mode has been implemented), but I'm also using it for an Interruption functionality and it serves its purpose nicely.

I also had/have some work on MidEffort but I thought no-one used this functionality, so it's been held 😄

---

To answer your question, KokoroSharp DOES support interruptions, but the OnProgress functionality is currently tied to KokoroTTS class and not exposed to KokoroWavSynthesizer.

On the bright side, PlaybackHandle exposes the Samples directly! So if you switch to that, you'll be able to turn them into WAV instead of 'Playing' them. Not sure if it's a better option than making KokoroWavSynthesizer a high-level utility class built on-top of KokoroTTS, though. Open to discuss, let me know any thoughts.

[rwfsmith]

How would I access those samples as they're generated so they can be streamed to a separate device?

[Lyrcaxis]

@rwfsmith caprimotion just posted a snippet that does this exact thing a couple days ago ^^
Give it a read: #48 (comment)