A FastAPI-based web service for forced alignment of audio and text using Montreal Forced Aligner (MFA) and Kalpy.

• RESTful API for audio-transcript alignment
• Returns alignment results as JSON with word and phone-level timing
• Currently hardcodes the dictionaries, so only supports one language at a time

• Anaconda or Miniconda

1. Clone this repository:

   git clone git@github.com:stephenmac7/mfa-service.git
   cd mfa-service

2. Create the conda environment from the environment file:

   conda env create -f environment.yml

3. Activate the environment:

   conda activate mfa-service

4. Download the required MFA models:

   mfa model download acoustic english_mfa
   mfa model download dictionary english_us_mfa

1. Activate the conda environment (if not already activated):

   conda activate mfa-service

2. Start the FastAPI server:

   fastapi dev main.py --port 8001

3. The API will be available at http://localhost:8001

   • Interactive API docs: http://localhost:8001/docs
   • Alternative docs: http://localhost:8001/redoc

POST /align

Upload a WAV file and transcript to get alignment results.

Parameters:

• audio (file): WAV audio file
• transcript (form field): Text transcript of the audio

Example using curl:

curl -X POST "http://localhost:8001/align" \
  -F "audio=@your_audio.wav" \
  -F "transcript=Your transcript text here"

Example using Python:

import requests

files = {'audio': open('your_audio.wav', 'rb')}
data = {'transcript': 'Your transcript text here'}
response = requests.post('http://localhost:8001/align', files=files, data=data)
print(response.json())

If you add new dependencies, update the environment.yml file and recreate the environment:

conda env update -f environment.yml --prune