Trim and split

Would it make sense to add a way of trimming silence when loading audio? Currently I’m doing this silliness, which works alright aside from being performance intensive, but the onset detector surely isn’t built for detecting offsets like this, I’d assume.

path = ... # Path to an audio file with a bunch of leading and trailing silence.
y, sr = librosa.load(path, sr=None)
onsets = librosa.frames_to_time(librosa.onset.onset_detect(y, sr), sr)
offsets = librosa.frames_to_time(librosa.onset.onset_detect(y[::-1], sr), sr)  # Suboptimal!
leading_silence = onsets[0]
trailing_silence = offsets[0]
total_duration = librosa.get_duration(y, sr)
y, _ = librosa.load(path, sr, offset=leading_silence, duration=total_duration-leading_silence-trailing_silence)

Also, perhaps if loading with a set duration longer than the actual file, then there should be the option of padding the sequence to the set length, or at least provide a user warning that less than duration was loaded.

[dpwe]

We could have trim accept a dB threshold input, and a flag to choose between interpreting that relative to the largest energy in the file, or full-scale. Maybe default to 60 dB below full-scale (rms sample value < 32 in 16 bit ints). It accepts n_fft (in lieu of window_length) and hop_length like the underlying rmse.

(I just noticed that rmse uses spectrogram under the hood. That’s quite extravagant compared to simply summing squared sample values.)

DAn.

On Mon, May 16, 2016 at 1:32 PM, Carl Thomé [e-mail] wrote:

A trim function would be very handy. [image: 👍]

— You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/librosa/librosa/issues/360#issuecomment-219489711

[carlthome]

On a similar note (https://github.com/librosa/librosa/issues/373) I think it would be neat to also have a padding built into load() when duration and offset exceed the audio file’s duration. Perhaps together with a warning in the weirder case when offset is larger than the duration.