Matthijs Hollemans

Add this line: ``` let elapsed = CACurrentMediaTime() - startTimes.remove(at: 0) ``` to `requestDidComplete()` to get the best possible estimate.

@anuragshas: > I will be helping with TextDecoderPrenet / TextDecoderPostnet Great! Are you also interested in looking at the tokenizer, since I believe the text pre- and post-net need to...

The `SpeechEncoderPrenet` is complete now. It gives the same results as the original model. However, there still are some TODOs in this part of the code to look at later.

The encoder is complete and verified to work (although there are some parts that possibly could be rewritten, marked with `TODO`). I've started adding the decoder but this doesn't work...

Thanks for the in-depth review, @sanchit-gandhi! > Wondering if it would have been better to copy everything from Speech2Text, including for the encoder? I know I pointed you in the...

If we're going to go this route, I'd rather get rid of the `feature` argument altogether and not have two ways to achieve the same thing. But that's something we...

One thing I want to do first is implement the `with-past` logic. Apparently this was one of the reasons for adding `feature` and it will affect some of the decisions...

Often you can ignore these version warnings but I guess this time they weren't kidding. :-) Glad to hear you got it solved. Another thing you can try is `pip...

Hi, I'm currently adding support for encoder-decoder models to the Core ML exporter. In theory this should work already but there are some issues with coremltools that prevent these kinds...

Yes, you're correct. But this is also an old model so it's not really worth fixing. I'd recommend getting a more up-to-date checkpoint from the TensorFlow models repo instead, and...