https://github.com/nalbion/whisper-server

streaming speech to text server using Whisper
https://github.com/nalbion/whisper-server

idiolect nlp whisper

Last synced: about 2 months ago
JSON representation

streaming speech to text server using Whisper

• Host: GitHub
• URL: https://github.com/nalbion/whisper-server
• Owner: nalbion
• Created: 2023-05-11T18:53:20.000Z (over 1 year ago)
• Default Branch: master
• Last Pushed: 2023-06-02T01:21:45.000Z (over 1 year ago)
• Last Synced: 2024-06-17T00:12:36.263Z (3 months ago)
• Topics: idiolect, nlp, whisper
• Language: Python
• Homepage:
• Size: 80.1 KB
• Stars: 71
• Watchers: 5
• Forks: 5
• Open Issues: 3
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Whisper Server

`whisper_server` listens for speech on the microphone and provides the results in real-time over 

[Server Sent Events](https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events/Using_server-sent_events) 

or [gRPC](https://grpc.io/).

This is intended as a local single-user server so that non-Python programs can use Whisper.

Integrates with the official [Open AI Whisper API](https://openai.com/research/whisper) and also

[faster-whisper](https://github.com/guillaumekln/faster-whisper).

## Requirements

- Python 3.10 or greater

- `pip install -r requirements.txt`

or [Docker](#running-in-docker)

## Running Whisper Server

```bash

python -m whisper_server

```

### Options

```

  -h, --help            show this help message and exit

  --whisper_impl {OpenAI,Faster Whisper}

                        Which Whisper implementation to use

  --model {tiny,base,small,medium,large}

                        Name of the Whisper model to use (without the language suffix)

  --language {all,en}   Specify 'en' if only English transcription is required

  --device {cpu,gpu}    Device to use for PyTorch inference

  --task {transcribe,translate}

                        Whether to perform speech recognition ('transcribe') or X->English translation

  --logging {error,warning,info,debug}

  --ssdp, --no-ssdp     '--ssdp' to allow discoverability of host IP by SSDP (default: False)

  --http, --no-http     '--http' to run HTTP server (default: False)

  --http_port HTTP_PORT

                        By default runs gRPC server on port 8080

  --grpc, --no-grpc     '--no-grpc false' to disable gRPC server (default: True)

  --grpc_port GRPC_PORT

                        By default runs gRPC server on port 9090

  --localhost, --no-localhost

                        '--no-localhost' if you need to access the server from another computer (default: True)

```

### Running in Docker

Build the Docker image:

```bash

docker build -t nalbion/whisper_server .

```

On Linux:

```bash

docker run --rm --device /dev/snd nalbion/whisper_server

```

On Windows use `wsl` (`PULSE_SERVER` is obtained by running `wsl echo $PULSE_SERVER`) [(credit)](https://stackoverflow.com/a/68316880/1225993):

__Note: This _should_ work - it does for `Dockerfile.cpp` - if you figure it out, let me know__ 

```bash

wsl docker run --rm -e "PULSE_SERVER=/mnt/wslg/PulseServer" -v /mnt/wslg/:/mnt/wslg/ nalbion/whisper_server

```

## Development

### Adding dependencies

Add any direct dependencies to `requirements.in` and run:

```bash

python -m piptools compile requirements.in --resolver=backtracking

```

### Generate gRPC code

```bash

python -m grpc_tools.protoc -I./whisper_server/proto --python_out=./whisper_server/proto --pyi_out=./whisper_server/proto --grpc_python_out=./whisper_server/proto ./whisper_server/proto/whisper_server.proto

sed -i "s/import whisper_server_pb2 as whisper__server__pb2/from whisper_server.proto import whisper_server_pb2 as whisper__server__pb2/" whisper_server/proto/whisper_server_pb2_grpc.py 

```