Clap processor: remove wasteful np.stack operations

[m-bain]

What does this PR do?

Upon profiling, it showed some strange result that the ClapProcessor was taking 0.5s to apply _get_input_mel(...) on short audio (less than 10s), whereas medium length audio (10s-20s) was taking only 0.02s

As it turns out there was a wasteful np.stack operation on the 1-D waveform numpy array, meaning that the 1-D array is unpacked then stacked back together again, with no effect. This PR removes this wasteful op and short audio is now also processed in 0.02s

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

@ArthurZucker
@sanchit-gandhi

[amyeroberts]

Thanks for updating @m-bain!

Could you provide an example snippet you used to run this for future reference of anyone visiting this PR?

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[m-bain]

@amyeroberts hows this ?

import time
import numpy as np

waveform = np.random.rand(100_000)
n_repeat = 10

t1_p = time.time()
prev_impl = np.stack(np.tile(waveform, n_repeat))
t2_p = time.time()

t1_n = time.time()
new_impl = np.tile(waveform, n_repeat)
t2_n = time.time()

assert (prev_impl == new_impl).all()
print(f"Time to process [prev. impl.]: {t2_p-t1_p:.3f}s")
print(f"Time to process [new. impl.]: {t2_n-t1_n:.3f}s")

Time to process [prev. impl.]: 0.883s
Time to process [new. impl.]: 0.001s

[amyeroberts]

@m-bain Thanks!

[ArthurZucker]

Thanks for the catch! 🤗

[sanchit-gandhi]

Awesome - thanks @m-bain! Also cc @ylacombe