Capisoft AI Transcribe App Monorepo

Speech-to-text API and clients using faster-whisper (SYSTRAN) for transcription and translation. Production-ready monorepo with clean architecture:

API:

• apps/api: FastAPI backend for Whisper transcription and translation.

Libraries:

• libraries/FastWhisper.Client (submodule): .NET client library for API integration.

Examples:

• examples/console-ui: Python console recorder that sends audio to API.
• examples/maui: .NET MAUI client (Windows, Android, iOS, MacCatalyst).

Infrastructure:

• infra/docker: Container build and compose files for API deployment.

Repository Layout

apps/
  api/
    app/
    requirements-api.txt
    scripts/install.ps1
libraries/
  FastWhisper.Client/  (submodule → capisoft-lib/FastWhisperApi.Client)
    src/FastWhisper.Client/
    examples/FastWhisper.Client.Example/
examples/
  console-ui/
    record_and_transcribe.py
    requirements-recorder.txt
  maui/
    Capisoft.AI.TranscribeApp/
infra/
  docker/
    Dockerfile
    docker-compose.yml
    requirements-docker.txt

Quick Start

1) API (local Python)

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install torch==2.6.* torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124
# Windows + CUDA 12.4: optional Flash Attention 2 wheel (faster GPU inference)
# Wheel is not in this repo; get it from Hugging Face (see link below), or:
pip install "https://huggingface.co/lldacing/flash-attention-windows-wheel/resolve/main/flash_attn-2.7.4%2Bcu124torch2.6.0cxx11abiFALSE-cp311-cp311-win_amd64.whl"
pip install -r apps/api/requirements-api.txt
uvicorn app.main:app --host 0.0.0.0 --port 8000 --app-dir apps/api

Pre-built Flash Attention 2 wheels for Windows (Python 3.11, CUDA 12.4, PyTorch 2.6, etc.) are available at lldacing/flash-attention-windows-wheel. Without this wheel, the API uses PyTorch SDPA.

For full API usage, parameters, and examples, see apps/api/README-API.md.

2) Console UI Example

.\.venv\Scripts\Activate.ps1
pip install -r examples/console-ui/requirements-recorder.txt
python examples/console-ui/record_and_transcribe.py --url http://127.0.0.1:8000

3) C# Client Library

The C# client is maintained in a separate repository as a submodule. Initialize it first:

git submodule update --init --recursive

Build and run the example:

dotnet build libraries/FastWhisper.Client/FastWhisper.Client.slnx
dotnet run --project libraries/FastWhisper.Client/examples/FastWhisper.Client.Example/FastWhisper.Client.Example.csproj

For detailed usage, see the FastWhisper.Client repository.

Install via NuGet:

dotnet add package FastWhisper.Client

Links:

• 📦 NuGet Package
• 📖 Documentation
• 🔧 Source Code

4) MAUI App Example

Open:

• examples/maui/Capisoft.AI.TranscribeApp/FastWhisperMaui.sln

Build targets:

• Windows: net10.0-windows10.0.19041.0 ✅ Tested
• Android: net10.0-android ✅ Tested
• iOS: net10.0-ios ⚠️ Not tested
• MacCatalyst: net10.0-maccatalyst ⚠️ Not tested

5) Docker API

Quick Start with Docker Hub Image

Use the pre-built base image from Docker Hub:

docker pull capitaine/fast-whisper-api:latest

Or run with Docker Compose:

docker compose -f infra/docker/docker-compose.yml up --build

For more information about the Docker deployment and base image, see infra/docker/README.md.

Docker Hub: capitaine/fast-whisper-api

Performance

Tested on NVIDIA GPUs 3060 Ti, 3080, and 3090 with very good performance thanks to the faster-whisper engine.

Environment

Copy .env.example to .env and set values as needed.

Credits

• faster-whisper — Fast Whisper inference (SYSTRAN).

Security and Contribution

• Security policy: SECURITY.md
• Contribution guide: CONTRIBUTING.md
• Changelog: CHANGELOG.md