Whisper_auto2lrc

English | 中文

Whisper_auto2lrc is a tool that uses the whisper model and a Python program to convert all audio files in a folder (and its subfolders) into .lrc subtitle files. If an lrc subtitle file already exists, it will be automatically skipped.

Now integrated with Faster Whisper for faster inference and lower memory usage! (Approximately 4 times faster inference speed and half the memory requirement)

Recommended to use Faster Whisper

What can this program do?

It uses the whisper speech-to-text model released by OpenAI. Users can specify the audio file folder to be processed, and the program will automatically recursively search for all audio files in the folder and transcribe them into lrc format subtitle files.

It is suitable for users who need to transcribe a large number of audio files into subtitles.

Automatic translation will be added in the future.

Screenshots

Using Faster Whisper

How to install

First, install Python, which has been tested on Python 3.10.11.

Then, install the Python dependencies required by whisper openai/whisper:

pip install git+https://github.com/openai/whisper.git 

If you want to use Faster Whisper, you need to install the dependencies of the Faster Whisper command line version Softcatala/whisper-ctranslate2

pip install -U whisper-ctranslate2

How to use

Using the source code program

Pull this repo and install the Python dependencies:

git clone https://github.com/bai0012/Whisper_auto2lrc

cd Whisper_auto2lrc

pip install -r requirements.txt 

With the terminal at the base of the project folder, run

python main.py

If you want to use Faster Whisper, run

python main_faster-whisper.py

(Faster Whisper has the Silero VAD filter turned on by default to skip blank audio segments over two seconds, and uses colors to indicate the confidence level of each character, with red indicating low confidence and green indicating high confidence)

In the window, select the folder path, choose the model size to use, enter the language of the audio files to be processed, and click start.

Troubleshooting

What's the difference between Faster Whisper and the original Whisper?

Faster-whisper is a reimplementation of OpenAI's Whisper model using CTranslate2, a fast inference engine for Transformer models.

This implementation is 4 times faster than openai/whisper, uses less memory and VRAM, and has the same accuracy.

According to guillaumekln/faster-whisper, the performance improvement of using the large-v2 model on NVIDIA Tesla V100S is shown in the following table:

Implementation	Precision	Beam size	Time required	Maximum GPU memory required	Maximum memory required
openai/whisper	fp16	5	4m30s	11325MB	9439MB
faster-whisper	fp16	5	54s	4755MB	3244MB
faster-whisper	int8	5	59s	3091MB	3117MB

Why is my Whisper only using the CPU?

Whisper can call GPUs that support CUDA. If you are sure that your GPU has been correctly installed and supports CUDA, try the following steps:

pip uninstall torch

pip cache purge

Then install the latest PyTorch using the command on the PyTorch official website

For example:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

You should be able to call the graphics card normally.

Why is ffmpeg giving an error when I use it?

You can try the following steps to reinstall ffmpeg:

pip uninstall ffmpeg

pip uninstall ffmpeg-python

pip install ffmpeg-python

Which model should I choose and what are the differences between each model?

Refer to openai/whisper

Size	Parameters	English-only model	Multilingual model	Required VRAM	Relative speed
tiny	39 M	tiny.en	tiny	~1 GB	~32x
base	74 M	base.en	base	~1 GB	~16x
small	244 M	small.en	small	~2 GB	~6x
medium	769 M	medium.en	medium	~5 GB	~2x
large	1550 M	N/A	large	~10 GB	1x

Choose the model based on the size of the VRAM on your GPU. Generally, the larger the model, the faster the speed and the lower the error rate.

(For example, if you have an RTX3060 12G, you can choose the large model, but if you have a GTX 1050ti 4G, you can only use the small model)

Faster Whisper uses the same models but requires less VRAM, so try it out for yourself.

(Faster Whisper can even run the large model on a 1050ti 4G, although it sometimes runs out of VRAM, so a 6G VRAM should be enough to run the large model)

What should I enter in the language input box?

Whisper supports speech-to-text in multiple languages, commonly used ones include:

Language	Code
Chinese	zh
English	en
Russian	ru
Japanese	ja
Korean	ko
German	de
Italian	it

For more languages, refer to: whisper/tokenizer.py

Star History

(This program was completed under the guidance of GPT 4)