Search results

I've noticed the way you split lines makes a difference with DeepL. Translating each line separately removes context, but running them together makes it blend the lines and repeat itself. See avatarthe's tests: The best fix I've found is putting quotes between the lines. 「」 and "" work about...

The 413 error should be fixed now. It'll also save the original Japanese output in a separate file when using DeepL (not possible when Whisper translates it). Vosk is based on Kaldi. It uses an acoustic model that looks at each fraction of a second, and predicts individual phonemes. Then a...

It's how many seconds the VAD waits before splitting the audio. Each chunk goes through Whisper separately, which prevents it from getting stuck on one line for the whole video.

I've updated the notebook. It should be much less likely to output things like "Please subscribe" now. For the rest of the video, or just one scene? You could try lowering the chunk_threshold to 2.0 or 1.0 if that happens.

I've written a notebook that combines Whisper with a separate VAD. It works much better than Whisper alone on long-form inputs, and also runs about 2-4x faster. https://colab.research.google.com/github/ANonEntity/WhisperWithVAD/blob/main/WhisperWithVAD.ipynb It's still far from perfect, though...

A big hurdle towards fine-tuning Whisper (or any other model) is a lack of Japanese training data. OpenAI's dataset has 15,914 hours of JP audio (7054 hours with Japanese transcripts, 8860 with English ones), and even that dwarfs the publicly available ones I'm aware of. Is anyone interested in...