❓ vad bad performance on child speech?

[garymmi]

I have a 16k audio file of child speech
after vad, I got
start=0.875, dur=1.375
as picture

I used model='silero_vad' , call the api
speech_timestamps = get_speech_ts(wav_data, model, num_steps=4)

here is the audio file
https://drive.google.com/file/d/1yivG8OE77TyfJE_KL2IYl-Ltm5cFujnX/view?usp=sharing

could you help me to see what'wrong or teach me how to fine tune the parameters ?
thanks

[snakers4]

Several remarks here

• We do not have a lot of children's speech per se in the training data
• This sample just looks too short to be cut by a VAD
• We were planning to release an adaptive post-processing tweak that would make parameter tuning obsolete
• You can use this snippet to tune params, albeit it should be done on longer audio files, becase this one is too short to be meaninful imo

speech_timestamps = get_speech_ts(wav, model,
                                  num_samples_per_window=2000,
                                  num_steps=8,
                                  visualize_probs=True)

This snippet produces this image

[snakers4]

Also tuning this maye be a good idea

min_speech_samples - minimum speech chunk duration in samples
min_silence_samples - minimum silence duration in samples between to separate speech chunks

[garymmi]

I got it, thank you very much