Add Speech to text notebook 211

[Debskij]

No description provided.

[ryanloney]

This notebook does not follow the contribution guide. Please review the guide. We need things like a header/title and layout the is similar to other notebooks.

A few other points:

• there should be a way to preview the audio clip by pressing play (should be doable with ipython)
• audio input type is too strict, need to be more flexible (more context below)
• can .ogg or other formats be supported?
• please do not use acronyms like CTC in headings
• please follow case style for headers and sub headings that matches other notebooks

Is there any way for us to take ANY .wav file and make sure it meets these requirements? (Without adding additional dependencies to the requirements.txt?)

assert sample_width == 2, "Only 16-bit WAV PCM supported"
assert compression_type == 'NONE', "Only linear PCM WAV files supported"
assert channel_num == 1, "Only mono WAV PCM supported"
assert sampling_rate == 16000, "Only 16 KHz audio supported"

librosa seems to have the ability to load OGG and MP3. Can we use it for this? https://librosa.org/doc/main/generated/librosa.load.html

[ryanloney]

Please make the following changes before approval to merge

[ryanloney]

@Debskij
Please replace the .wav sample file with edge_to_cloud.ogg (in the .zip file below). Then it is good to merge.
edge_to_cloud.zip

[ryanloney]

Suggested change

	"### Run Decoding and Print Output."
	"### Run Decoding and Print Output"

[ryanloney]

Suggested change

	"### Do Inference!\n",
	"### Do Inference\n",

[ryanloney]

@Debskij please clear the outputs from Jupyter cells and commit clean version.

[ryanloney]

@helena-intel I'm fine with merging this tomorrow, as long as you are

[ryanloney]

Just for fun, I tried transcribing a CSPAN video (public domain) to see how long it would take. This is a 17 minute video and the model was able to process on my laptop's Tiger Lake CPU in 4s and only 2s on iGPU. https://youtu.be/Wp-WiNXH6hI

This is the output. I'm impressed.
cspan.txt

[helena-intel]

Thanks @Debskij ! This is a great notebook - so many possibilities. I love that this allows you to run speech-to-text for private data that you do not want to upload to some cloud server.

I will approve and merge this now. I have one non-blocking change request: there is a DeprecationWarning about waveplot. The replacement (according to docstring) is waveshow. I tried to simply replace the method but that did not work. Can you make a separate PR that changes waveplot to waveshow?