Meeting Audio -> Transcribe & TTS (Flask demo)

This demo provides a local web UI to:

• Upload meeting audio (mp3/wav)
• Transcribe using Faster Whisper
• Translate segments (ja<->en) using Helsinki Marian models
• Synthesize speech using the local NeuTTSAir implementation in lib/neuttsair

This repository includes a small single-page web UI (Flask + SSE) that shows live progress and lets you download the generated audio.

Quick overview

• Uploads are stored at uploads/<task_uuid>/<original_filename>.
• Outputs are written to outputs/<task_uuid>/<original_filename>.
• Models are loaded at process start (see app.py). Progress state is stored in-memory (not durable).

Requirements

• Python 3.10+ recommended
• Install top-level requirements:

pip install -r requirements.txt

• Install lib/neuttsair additional requirements:

pip install -r lib\neuttsair\requirements.txt

Install eSpeak (required for some phonemizers)

Some TTS phonemizers require eSpeak (espeak-ng). Follow the platform instructions in the eSpeak NG guide:

eSpeak NG installation guide

On Windows you may also need to set environment variables (example):

set PHONEMIZER_ESPEAK_LIBRARY=C:\Program Files\eSpeak NG\libespeak-ng.dll
set PHONEMIZER_ESPEAK_PATH=C:\Program Files\eSpeak NG

Configuration (app.py constants)

At the top of app.py you can customize model names and OS-specific phonemizer paths. Defaults included in the file:

WHISPER_MODEL_SIZE = "base"  # smaller model for CPU usage
JA_EN_MODEL_NAME = "Helsinki-NLP/opus-mt-ja-en"
EN_JA_MODEL_NAME = "Helsinki-NLP/opus-mt-en-jap"

WINDOWS_PHONEMIZER_ESPEAK_LIBRARY = "c:\\Program Files\\eSpeak NG\\libespeak-ng.dll"
WINDOWS_PHONEMIZER_ESPEAK_PATH = "c:\\Program Files\\eSpeak NG"

MACOS_PHONEMIZER_ESPEAK_LIBRARY = "/opt/homebrew/Cellar/espeak/1.48.04_1/lib/libespeak.1.1.48.dylib"

Adjust these values if you change install locations or model choices.

Run (Windows cmd.exe)

1. Create and activate a venv (recommended):

python -m venv .venv
.venv\Scripts\activate

1. Install requirements (both top-level and lib/neuttsair):

pip install -r requirements.txt
pip install -r lib\neuttsair\requirements.txt

1. Start the web UI:

python app.py

Open http://127.0.0.1:5000/ in your browser. Drag & drop audio files or pick files to upload. The SPA shows live progress via SSE and provides download links when processing completes.

Notes & caveats

• The demo loads models on startup and runs on CPU; expect high memory and CPU usage for larger models.
• lib/neuttsair/neutts.py provides the local NeuTTSAir API. If you modify it, update lib/neuttsair/requirements.txt as needed.
• Progress state is in-memory (progress_state dict). For production use, switch to a durable store (Redis) and a task queue (RQ/Celery).

Development tips

• If you get errors about missing library paths for eSpeak, check the PHONEMIZER_ESPEAK_LIBRARY and PHONEMIZER_ESPEAK_PATH variables in app.py and ensure the file exists.
• To reduce memory usage, consider loading smaller Whisper models or using an external model server.