alqalign

KO | EN

gitlite — search

#javascript #react #llm #docker #nodejs #macos #awesome #kubernetes #vue #pytorch #security #ios #openai #open-source

(★ 78)

multilingual speech aligner

File Explorer

Download Latest Version (.zip)

- - python.yml
- - __init__.py
  - dataset.py
  - decoder.py
  - loader.py
  - model.py
  - recognizer.py
- - download_model.py
- - __init__.py
  - ctc_segmentation.py
  - ctc_segmentation_dyn.pyx
  - partitioning.py
- - __init__.py
  - checkpoint_utils.py
  - phonetics.py
  - tensor.py
  - text.py
- __init__.py
- app.py
- audio.py
- config.py
- eval.py
- model.py
- process_alignment.py
- process_audio.py
- process_text.py
- record.py
- run.py
- run_step.py
- instruction.md
- language.md
- - utt.txt
  - utt.wav
- - utt.txt
  - utt.wav
- - utt.txt
  - utt.wav
- test_alignment.py
.gitignore
LICENSE
README.md
requirements.txt
setup.py

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# ALQAlign

![CI Test](https://github.com/xinjli/alqalign/actions/workflows/python.yml/badge.svg)

alqalign is a phoneme-based multilingual speech alignment toolkit.

It is supposed to be able to handle ~8k language (at least theoretically). See the [full list](./doc/language.md) of supported languages.

## Install

```bash
python setup.py install
```

## Usage

The basic command-line usage with main configures is as follows:

For the details of usage, check the [instruction page](./doc/instruction.md).

```bash
python -m alqalign.run  --lang   <your target language> 
                        --audio  <path to your audio file> 
                        --text   <path to your text file> 
                        --output <path to an output directory>
```

You can also use it directly as follows:

```python
In [1]: from alqalign.app import align

In [2]: align('./samples/eng/utt.wav', './samples/eng/utt.txt', 'eng')
Out[2]:
[{'utt_id': 'utt-00000-0000028-0000367',
  'start': 0.28,
  'end': 3.67,
  'text': 'A PROGRAMMER WALKS TO THE BUTCHER SHOP AND BUYS A KILO OF MEAT.',
  'score': -0.14},
 {'utt_id': 'utt-00001-0000384-0000917',
  'start': 3.85,
  'end': 9.17,
  'text': 'AN HOUR LATER HE COMES BACK UPSET THAT THE BUTCHER SHORTCHANGED HIM BY 24 GRAMS.',
  'score': -0.13}]
```

## Tutorial

There is one English sample in the `samples/eng` directory. It contains two files:

- `utt.wav`: a wav file containing 10 seconds of speech.
- `utt.txt`: a text file containing two lines as follows:

```text
A programmer walks to the butcher shop and buys a kilo of meat.
An hour later he comes back upset that the butcher shortchanged him by 24 grams.
```

To apply the alignment for each line, you can run the following command:

```
python -m alqalign.run --lang=eng --audio=./samples/eng/utt.wav --text=./samples/eng/utt.txt --output=./samples/output/eng
```

The output will be in the `./samples/output/eng` directory. It contains a few files including

- `segments`: containing timestamps
- `text`: containing the aligned text

In this sample, the `segments` is

```text
utt-00000-0000028-0000367 utt 0.28 3.67
utt-00001-0000384-0000917 utt 3.85 9.17
```

and `text` is

```text
utt-00000-0000028-0000367 A PROGRAMMER WALKS TO THE BUTCHER SHOP AND BUYS A KILO OF MEAT.
utt-00001-0000384-0000917 AN HOUR LATER HE COMES BACK UPSET THAT THE BUTCHER SHORTCHANGED HIM BY 24 GRAMS.
```

alqalign

File Explorer

Today's picks

Was this content helpful?

Recent Feedback

alqalign

File Explorer

Today's picks

Was this content helpful?

Recent Feedback

Download README