Merge pull request #244 from ahmetoner/upgrade-whisper-v20240930

Upgrade OpenAI Whisper to v20240930 (turbo)

CHANGELOG.md

@@ -4,6 +4,15 @@ Changelog
 Unreleased
 ----------
 
+### Changed
+
+- Upgraded
+  - [openai/whisper](https://github.com/openai/whisper)@[v20240930](https://github.com/openai/whisper/releases/tag/v20240930)
+  - fastapi to v0.115.0
+  - uvicorn to v0.31.0
+  - tqdm to v4.66.5
+  - python-multipart to v0.0.12
+
 [1.5.0] (2024-07-04)
 --------------------
 

Dockerfile

@@ -63,4 +63,4 @@ RUN poetry install
 
 EXPOSE 9000
 
-ENTRYPOINT ["gunicorn", "--bind", "0.0.0.0:9000", "--workers", "1", "--timeout", "0", "app.webservice:app", "-k", "uvicorn.workers.UvicornWorker"]
+ENTRYPOINT ["whisper-asr-webservice"]

Dockerfile.gpu

@@ -81,4 +81,4 @@ RUN $POETRY_VENV/bin/pip install torch==1.13.1+cu117 -f https://download.pytorch
 
 EXPOSE 9000
 
-CMD gunicorn --bind 0.0.0.0:9000 --workers 1 --timeout 0 app.webservice:app -k uvicorn.workers.UvicornWorker
+CMD whisper-asr-webservice

README.md

@@ -11,7 +11,7 @@ Whisper is a general-purpose speech recognition model. It is trained on a large
 
 Current release (v1.5.0) supports following whisper models:
 
-- [openai/whisper](https://github.com/openai/whisper)@[v20231117](https://github.com/openai/whisper/releases/tag/v20231117)
+- [openai/whisper](https://github.com/openai/whisper)@[v20240930](https://github.com/openai/whisper/releases/tag/v20240930)
 - [SYSTRAN/faster-whisper](https://github.com/SYSTRAN/faster-whisper)@[v1.0.3](https://github.com/SYSTRAN/faster-whisper/releases/tag/1.0.3)
 
 ## Quick Usage

app/webservice.py

 import importlib.metadata
 import os
 from os import path
-from typing import Annotated, BinaryIO, Union
+from typing import Annotated, BinaryIO, Optional, Union
 from urllib.parse import quote
 
+import click
 import ffmpeg
 import numpy as np
+import uvicorn
 from fastapi import FastAPI, File, Query, UploadFile, applications
 from fastapi.openapi.docs import get_swagger_ui_html
 from fastapi.responses import RedirectResponse, StreamingResponse
@@ -14,9 +16,9 @@ from whisper import tokenizer
 
 ASR_ENGINE = os.getenv("ASR_ENGINE", "openai_whisper")
 if ASR_ENGINE == "faster_whisper":
-    from .faster_whisper.core import language_detection, transcribe
+    from app.faster_whisper.core import language_detection, transcribe
 else:
-    from .openai_whisper.core import language_detection, transcribe
+    from app.openai_whisper.core import language_detection, transcribe
 
 SAMPLE_RATE = 16000
 LANGUAGE_CODES = sorted(tokenizer.LANGUAGES.keys())
@@ -122,3 +124,28 @@ def load_audio(file: BinaryIO, encode=True, sr: int = SAMPLE_RATE):
         out = file.read()
 
     return np.frombuffer(out, np.int16).flatten().astype(np.float32) / 32768.0
+
+@click.command()
+@click.option(
+    "-h",
+    "--host",
+    metavar="HOST",
+    default="0.0.0.0",
+    help="Host for the webservice (default: 0.0.0.0)",
+)
+@click.option(
+    "-p",
+    "--port",
+    metavar="PORT",
+    default=9000,
+    help="Port for the webservice (default: 9000)",
+)
+@click.version_option(version=projectMetadata["Version"])
+def start(
+    host: str,
+    port: Optional[int] = None
+):
+    uvicorn.run(app, host=host, port=port)
+
+if __name__ == "__main__":
+    start()
\ No newline at end of file

docs/build.md

@@ -24,7 +24,7 @@ poetry install
 Starting the Webservice:
 
 ```sh
-poetry run gunicorn --bind 0.0.0.0:9000 --workers 1 --timeout 0 app.webservice:app -k uvicorn.workers.UvicornWorker
+poetry run whisper-asr-webservice --host 0.0.0.0 --port 9000
 ```
 
 ### Build

docs/environmental-variables.md

@@ -15,7 +15,7 @@
 export ASR_MODEL=base
 ```
 
-Available ASR_MODELs are `tiny`, `base`, `small`, `medium`, `large` (only OpenAI Whisper), `large-v1`, `large-v2` and `large-v3`.
+Available ASR_MODELs are `tiny`, `base`, `small`, `medium`, `large`, `large-v1`, `large-v2`, `large-v3`, `turbo`(only OpenAI Whisper) and `large-v3-turbo`(only OpenAI Whisper).
 
 For English-only applications, the `.en` models tend to perform better, especially for the `tiny.en` and `base.en` models. We observed that the difference becomes less significant for the `small.en` and `medium.en` models.
 

docs/index.md

@@ -4,7 +4,7 @@ Whisper is a general-purpose speech recognition model. It is trained on a large
 
 Current release (v1.5.0) supports following whisper models:
 
-- [openai/whisper](https://github.com/openai/whisper)@[v20231117](https://github.com/openai/whisper/releases/tag/v20231117)
+- [openai/whisper](https://github.com/openai/whisper)@[v20240930](https://github.com/openai/whisper/releases/tag/v20240930)
 - [SYSTRAN/faster-whisper](https://github.com/SYSTRAN/faster-whisper)@[v1.0.3](https://github.com/SYSTRAN/faster-whisper/releases/tag/1.0.3)
 
 ## Quick Usage

pyproject.toml

@@ -8,6 +8,9 @@ authors = ["Ahmet Öner", "Besim Alibegovic"]
 readme = "README.md"
 packages = [{ include = "app" }]
 
+[tool.poetry.scripts]
+whisper-asr-webservice = "app.webservice:start"
+
 [[tool.poetry.source]]
 name = "pytorch"
 url = "https://download.pytorch.org/whl/cpu"
@@ -15,15 +18,14 @@ priority = "explicit"
 
 [tool.poetry.dependencies]
 python = "^3.10"
-uvicorn = { extras = ["standard"], version = "^0.30.1" }
-gunicorn = "^22.0.0"
-tqdm = "^4.66.4"
-python-multipart = "^0.0.9"
+uvicorn = { extras = ["standard"], version = "^0.31.0" }
+tqdm = "^4.66.5"
+python-multipart = "^0.0.12"
 ffmpeg-python = "^0.2.0"
-fastapi = "^0.111.0"
+fastapi = "^0.115.0"
 llvmlite = "^0.43.0"
 numba = "^0.60.0"
-openai-whisper = "^20231117"
+openai-whisper = "^20240930"
 faster-whisper = "^1.0.3"
 torch = [
   { markers = "sys_platform == 'darwin' and platform_machine == 'arm64'", url = "https://download.pytorch.org/whl/cpu/torch-1.13.1-cp310-none-macosx_11_0_arm64.whl" },