Because who likes reading.

I want the robot to TALK. My daughter has no patience for reading all those words every time.

Speechbrain

I tried Speechbrain. It’s not bad. A little flat though. On the plus side, it super fast and it runs on CPU. Super fast meaning that on my 2020 MacBook Pro it takes a bunch of seconds to render 20-30s of speech. Downsides:

• as I said, flat-ty
• pronounces wrong, sometimes (es Hermione)
• need to tweak the sentence. E.g. I needed to add a blank space at beginning + end of sentences, otherwise sometimes it would say the first word wrong (or the last one).
• doesn’t read numbers! (WHY?!?)
• just english, at this time (may 11 2023)

Some examples:

1.wav

5.wav

Here’s the problem. I tried the same block of text on elevenlabs and was blown away with the differences. The guys at elevenlabs are still my reference in TTS world.

Might pull the trigger and pay up for elevenlabs. It’s PRICEY.