whisper 🤫

Add support for OpenAI's whisper model through transformers ✔️

• this makes vid2cleantxt significantly more robust/useful for day-to-day cases
• the ability of the openAI models to translate does carry over to vid2cleantxt*! transcribe a Chinese video to English with model_id="openai/whisper-small"for example

* all testing here was done en-to-en, official testing/support for other languages to come later :)

• What's Changed

• Whisper by @pszemraj in #14

Full Changelog: v0.2...v0.2.5

package + python API

Finally, a python API to transcribe things instead of using a CLI or custom notebook!

See the example here on Colab.

high-level API

Install with pip:

pip install git+https://github.com/pszemraj/vid2cleantxt.git

Use in python:

import vid2cleantxt

text_output_dir, metadata_output_dir = vid2cleantxt.transcribe.transcribe_dir(
    input_dir="path/to/video/files",
    chunk_length=15,
)

---

What's Changed

• packaging by @JonathanLehner in #9
• Update v2ct_utils.py by @JonathanLehner in #10
• Update transcribe.py by @JonathanLehner in #11
• Add spacy workaround by @pszemraj in #12
• Update docs by @pszemraj in #13

New Contributors

• @JonathanLehner made their first contribution in #9

Full Changelog: v0.1.21...v0.2

v0.1.21 - neuspell workaround

• this version is a bug fix: in NeuSpell, there is a bug loading model via the recommended API, so added code to use SymSpell if that is the case automatically. Check the log file to see/confirm that this is happening when you transcribe.

What's Changed

• Add workaround for neuspell initialization bug by @pszemraj in #8

Full Changelog: v0.1.2...v0.1.21

v0.1.2

What's Changed

• User friendly by @pszemraj in #3
• Model updates by @pszemraj in #4
• Add support for the Hubert CTC model by @pszemraj in #5

Next Release

• Adding PDF generation from text files post-transcription.

Full Changelog: https://github.com/pszemraj/vid2cleantxt/commits/v0.1.2