Java example to remove uh/um filler words from a video

[siddhsql]

Hello,

Is there any example showing how one could use this library to process a video file and remove occurrences of filler words like uh/um etc.? is that possible?

[csukuangfj]

sorry, we don't support that.

[csukuangfj]

Please describe in detail what you want.

[siddhsql]

I want to

1. extract audio waveform
2. get an array of corresponding words and associated timestamps e.g., [(Hello, 0:00), (world, 0:05)...]
   in Java. is that possible? thanks.

[csukuangfj]

If you use a zipformer model, then it is possible.

please see

sherpa-onnx/sherpa-onnx/java-api/src/com/k2fsa/sherpa/onnx/OfflineRecognizerResult.java

Line 8 in ce71b63

private final float[] timestamps;

The timestamp contains start time of each token. It is your responsibility to get the timestamp for each word from tokens.

Once you have the timestamp info about each word, it should be straightforward to get the corresponding waveform.

[siddhsql]

thanks. you have multiple speech recognition backends. is there a recommended one to use?

[csukuangfj]

Please see
https://k2-fsa.github.io/sherpa/intro.html
for a comparison among k2-fsa/sherpa, k2-fsa/sherpa-ncnn, and k2-fsa/sherpa-onnx.

If your favorite programming language is Java, then I suggest that you use k2-fsa/sherpa-onnx