https://github.com/ventz/whisper-openai-container
OpenAI Whisper Container (GPU and CPU) and Lambda (CPU) - speech recognition model
https://github.com/ventz/whisper-openai-container
Last synced: 5 months ago
JSON representation
OpenAI Whisper Container (GPU and CPU) and Lambda (CPU) - speech recognition model
- Host: GitHub
- URL: https://github.com/ventz/whisper-openai-container
- Owner: ventz
- License: mit
- Created: 2023-05-03T18:43:12.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2024-05-09T00:22:51.000Z (over 2 years ago)
- Last Synced: 2026-01-18T15:22:11.073Z (7 months ago)
- Language: Shell
- Size: 5.13 MB
- Stars: 60
- Watchers: 1
- Forks: 8
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# What is this?
This is the OpenAI Whisper project (https://github.com/openai/whisper - Offline Speech Recognition model) - inside a Container, with an option to deploy as a stand-alone Docker container, or an AWS Lambda function that is container backed.
In summary: it lets you transcribe voice to text *extremely* accurately and quickly, for free.
# How Do I use this?
There are 2 ways to run/interact with this:
* As a "regular container" (Docker)
or
* As an AWS Lambda (container backed) function - via an a direct API, or S3 "put" automation.
## 1.) As a "regular container":
```
docker exec -it ventz/whisper /bin/bash"
```
```
# Assuming you have a 'recording.mp4' and have pulled it/mounted it on the container:
whisper 'recording.mp4' --language English --model base --fp16 False
```
## 2.) As an AWS Lambda (container backed) functions:
The idea is that you will setup a S3 bucket with a hook that calls
this Lambda when a new object is created or dropped.
This involves:
a.) Tagging the local docker image and pushing it to ECR:
```
docker tag ventz/whisper:latest 123456789012.dkr.ecr.us-east-1.amazonaws.com/whisper:latest
docker push 123456789012.dkr.ecr.us-east-1.amazonaws.com/whisper:latest
```
b.) Deploying a new Lambda function from ECR:
```
aws lambda create-function --region us-east-1 --function-name transcribe \
--package-type Image \
--code ImageUri= \
--role arn:aws:iam::123456789012:role/service-role/transcribe
```
NOTE: The role needs to have:
i.) AWSLambdaBasicExecutionRole (for: 'logs:CreateLogStream', and 'logs:PutLogEvents')
```
{
"Version": "2012-10-17",
"Statement": [
{
"Effect": "Allow",
"Action": "logs:CreateLogGroup",
"Resource": "arn:aws:logs:us-east-1:123456789012:*"
},
{
"Effect": "Allow",
"Action": [
"logs:CreateLogStream",
"logs:PutLogEvents"
],
"Resource": [
"arn:aws:logs:us-east-1:123456789012:log-group:/aws/lambda/transcribe:*"
]
}
]
}
```
and
ii.) Write access to S3 bucket:
```
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "VisualEditor0",
"Effect": "Allow",
"Action": [
"s3:PutObject",
"s3:GetObject",
"s3:ListBucket"
],
"Resource": "arn:aws:s3:::/*"
}
]
}
```
c.) Update code if you ever re-configure/re-build your container/Dockerfile:
```
# NOTE: This assumes your function was deployed with the name 'transcribe'
aws lambda update-function-code --function-name transcribe --image-uri $(aws lambda get-function --function-name transcribe | jq -r '.Code.ImageUri')
```
You can check when done with:
```
while [ "$(aws lambda get-function --function-name transcribe | jq -r '.Configuration.LastUpdateStatus')" != "Successful" ]; do
sleep 1
done
```
# Works locally but not in AWS Lambda?
The container has to be `amd64` due to the statically compiled ffmpeg
being only amd64. This means you cannot use the ARM64 Lambdas.
If you are building the container on a Mac M# series model and pushing to ECR, replace the 1st line in the Dockerfile with:
```
FROM --platform=linux/amd64 public.ecr.aws/lambda/python:3.12
```
### MANUALLY TESTING THE LAMBDA LOCALLY (not within AWS):
```
docker run -it --rm -d -p 9000:8080 --name whisper ventz/whisper
```
and then
```
curl -XPOST "http://localhost:9000/2015-03-31/functions/function/invocations" -d @test-s3-json
```
NOTE: This is a "fake" event just to make sure you can locally run
the lambda. You will need a real s3 bucket and real file/recording + IAM permissions(see test-s3-json)