Raden Muaz Munim
Raden Muaz Munim
was writing rnnt till I see this commit... btw how is the WER on dev set?
For MLPerf RNNT `This is not typical preprocessing, since it takes place as part of the model's measured runtime, not before the model runs.` https://github.com/mlcommons/inference/tree/master/speech_recognition/rnnt And for speech models in...
I have moved `tinygrad/nn/signal.py` to `examples/mlperf/audio_helpers.py`. Also added (log-) mel spectrogram layer.
Really wanted to try but don't have much free time. I'll see what I can do.