"Quint" is designed to enhance the podcast experience. It simplifies the process for users, making it easier for them to understand and navigate podcasts by providing concise summaries, highlights, and transcripts.
Below is a list of the core API endpoints offered by Quint:
Simply provide a YouTube video ID. Quint will fetch the video, extract its audio content, and return a transcription of the audio.
GET /youtube_transcript?video_id=YOUR_YOUTUBE_VIDEO_ID
Upload an audio file and instantly receive its transcription in text format.
POST /file_transcript
Submit a lengthy text and get it divided into semantically meaningful chunks or paragraphs.
POST /chunk
Submit a text and let Quint analyze it. The endpoint returns the index of the most descriptive sentence based on the embeddings.
POST /best_sentence
This project is licensed under the MIT License - see the LICENSE.md file for details.
Importand note: I highly reccomend using the JAX solution, as it is much faster than OpenAI proposed way. Please refer to this git Whisper Jax for more details. I will attach one of the table from this git repo:
Table 1: Average inference time in seconds for audio files of increasing length. GPU device is a single A100 40GB GPU. TPU device is a single TPU v4-8.
| OpenAI | Transformers | Whisper JAX | Whisper JAX | |
|---|---|---|---|---|
| Framework | PyTorch | PyTorch | JAX | JAX |
| Backend | GPU | GPU | GPU | TPU |
| 1 min | 13.8 | 4.54 | 1.72 | 0.45 |
| 10 min | 108.3 | 20.2 | 9.38 | 2.01 |
| 1 hour | 1001.0 | 126.1 | 75.3 | 13.8 |
Of course you are free to choose any cloud provider as Google/Azure/AWS. However, one of the issues for me is that thoose providers do not have ADA6000 GPU's which are cheaper then A100 and likes and provide the same compute capability. For that reason I propose - Datacrunch. This is not an advertisement, I'm in no way affiliated to them, but I found their service as best for cost/price for transcribtion.
- We need to create an account via Sign-Up
- Add card and top-up minumum ammount (20USD currently)
- In the account settings Generate credentials and fill in .env.sample with Id and Secret
- Rename .env.sample to .env
- Run direnv reload in terminal
import os
from datacrunch import DataCrunchClient
# Get client secret from environment variable
CLIENT_SECRET = os.environ['DATACRUNCH_CLIENT_SECRET']
CLIENT_ID = os.environ['DATACRUNCH_CLIENT_ID']
# Create datcrunch client
datacrunch = DataCrunchClient(CLIENT_ID, CLIENT_SECRET)import os
import subprocess
import shutil
email = os.environ['EMAIL']
key_filename = email.replace("@", "_").replace(".", "_")
def generate_ssh_key(email: str) -> str:
try:
# Create a filename based on the email
# Define the command and its arguments as a list
cmd = [
"ssh-keygen",
"-t", "ed25519",
"-C", email,
"-f", key_filename, # Using the modified filename
"-N", "" # Empty passphrase
]
# Run the command
process = subprocess.Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
# Send an empty passphrase (newline) twice: once for the passphrase itself and once to confirm it
process.communicate(input=b'\n\n')
if process.returncode != 0:
raise Exception("Error generating the key.")
# Read the public key
with open(f"{key_filename}.pub", "r") as f:
pubkey = f.read()
# Move generated keys to the ~/.ssh folder
shutil.move(key_filename, os.path.expanduser(f"~/.ssh/{key_filename}"))
shutil.move(f"{key_filename}.pub", os.path.expanduser(f"~/.ssh/{key_filename}.pub"))
print(f"SSH key for {email} generated successfully!")
return pubkey
except Exception as e:
print(f"Failed to generate SSH key for {email}. Reason: {e}")
return None
# Get all SSH keys
ssh_keys = datacrunch.ssh_keys.get()
ssh_keys = list(map(lambda key: key.id, ssh_keys))
if len(ssh_keys) == 0:
public_key = generate_ssh_key(email)
datacrunch.ssh_keys.create(email,public_key)
ssh_keys = datacrunch.ssh_keys.get()
ssh_keys = list(map(lambda key: key.id, ssh_keys))#We select accessible GPU's that work really fast
instance_type=''
if datacrunch.instances.is_available(instance_type='1A6000.10V'):
instance_type='1A6000.10V'
elif datacrunch.instances.is_available(instance_type='1A6000ADA.10V'):
instance_type='1A6000ADA.10V'Edit path to script according to enviroment you are in. Otherwise you can just run the notebook.
#Read the Bash script's contents
with open('../scripts/clone_github_code.sh', 'r') as file:
script_contents = file.read()
#Use the script_contents in the datacrunch function
datacrunch.startup_scripts.create('cuda_drivers', script_contents)
script_id = datacrunch.startup_scripts.get()[0].id#Create a new instance
instance = datacrunch.instances.create(instance_type=instance_type,
image='ubuntu-22.04-cuda-12.0-docker',
ssh_key_ids=ssh_keys,
hostname='quint',
description='Transcription and summarization API',
startup_script_id=script_id,
os_volume={
"name": "OS volume",
"size": 100
}
)
import time
while datacrunch.instances.get()[0].status=='provisioning':
time.sleep(5)
print('Instance is running, you can now connect to it.')host = datacrunch.instances.get()[0].ip
print(f'ssh root@{host} -i ~/.ssh/{key_filename}')
Copy this command and run in terminal to connect to the instance.#Connect to server via ssh and perform command "sudo reboot" to restart server for enabling Nvidia Drivers
sudo rebootInside of instance terminal run this commands to copy latest version
apt install gh
git clone https://github.com/poloniki/quint.gitBuild and run docker
docker build -t quint --file Dockerfile.jax .
docker run --gpus all -p 80:80 quintYour API endpoint would be avialable on the instance Ip-adress.