more flexibility on Audio Transcriber

[ohjho]

first of all, just want to say this is a great package! 💯 Thanks for putting this out there ❤️

on to the audio transcription issue...

• have you thought about using a "local" Transcriber like in openscenesense-ollama?
• or alternatively, loading whisper using the huggingface serverless inference api (e.g. see below)
• and an option to turn off audio transcription (which might be useful if the video has no audio track)

import requests

API_URL = "https://api-inference.huggingface.co/models/openai/whisper-small"
headers = {"Authorization": "Bearer hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"}

def query(filename):
    with open(filename, "rb") as f:
        data = f.read()
    response = requests.post(API_URL, headers=headers, data=data)
    return response.json()

output = query("sample1.flac")

[ymrohit]

Hi @ohjho , Thank you for your feedback and suggestion. I think the huggingface serverless inference api can be implemented and would help this library a lot along with the option to turn off the audio. I thought of adding local inference but I didn't want to make this library heavy with powerful dependencies. The main goal of this library is to be light weight and fast so that it can run on any device. I am open for further feedback, please feel free to let me know. Also please feel free to contribute any new ideas.