narabas

narabas: Japanese phoneme forced alignment tool

Input

Wav file (sample file: input.wav) and phoneme from pyopenjtalk

Output

Assigned phoneme and corresponding start and end times. Output for sample input wav file is below.

INFO narabas.py (192) : 0.220 0.300 a
INFO narabas.py (192) : 0.300 0.320 n
INFO narabas.py (192) : 0.320 0.400 e
INFO narabas.py (192) : 0.400 0.440 m
INFO narabas.py (192) : 0.440 0.480 u
INFO narabas.py (192) : 0.480 0.520 s
INFO narabas.py (192) : 0.520 0.580 u
INFO narabas.py (192) : 0.580 0.620 m
INFO narabas.py (192) : 0.620 0.680 e
INFO narabas.py (192) : 0.680 0.720 n
INFO narabas.py (192) : 0.720 0.860 o
INFO narabas.py (192) : 0.860 0.880 ts
INFO narabas.py (192) : 0.880 0.980 u
INFO narabas.py (192) : 0.980 1.060 i
INFO narabas.py (192) : 1.060 1.100 k
INFO narabas.py (192) : 1.100 1.160 o
INFO narabas.py (192) : 1.160 1.220 w
INFO narabas.py (192) : 1.220 1.620 a

Usage

For the sample input wav file,

python3 narabas.py

If you want to infer with onnxruntime,

python3 narabas.py --onnx

To specify input wav file,

python3 narabas.py --input <input wav file>

Reference

narabas

Framework

Pytorch

Model Format

ONNX opset=11

Netron

narabas-v0.onnx.prototxt