PyThaiASR

Python Thai Automatic Speech Recognition

PyThaiASR is a Python package for Automatic Speech Recognition with focus on Thai language. It have offline thai automatic speech recognition model.

License: Apache-2.0 License

Google Colab: Link Google colab

Model homepage: https://huggingface.co/airesearch/wav2vec2-large-xlsr-53-th

Install

pip install pythaiasr

For Wav2Vec2 with language model: if you want to use wannaphong/wav2vec2-large-xlsr-53-th-cv8-* model with language model, you needs to install by the step.

pip install pythaiasr[lm]
pip install https://github.com/kpu/kenlm/archive/refs/heads/master.zip

For live audio streaming: If you want to use live audio streaming from microphone/soundcard, you need to install PyAudio:

pip install pythaiasr[stream]

Usage

File-based ASR

from pythaiasr import asr

file = "a.wav"
print(asr(file))

Live Audio Streaming

Stream audio directly from your microphone/soundcard:

from pythaiasr import stream_asr

# Stream audio and print transcriptions in real-time
for transcription in stream_asr(chunk_duration=5.0):
    print(transcription)
    # Press Ctrl+C to stop

API

asr

asr(data: str, model: str = _model_name, lm: bool=False, device: str=None, sampling_rate: int=16_000)

• data: path of sound file or numpy array of the voice
• model: The ASR model
• lm: Use language model (except airesearch/wav2vec2-large-xlsr-53-th model)
• device: device
• sampling_rate: The sample rate
• return: thai text from ASR

stream_asr

stream_asr(model: str = _model_name, lm: bool=False, device: str=None, chunk_duration: float=5.0, sampling_rate: int=16_000)

• model: The ASR model
• lm: Use language model (except airesearch/wav2vec2-large-xlsr-53-th model)
• device: device
• chunk_duration: Duration of each audio chunk in seconds (default: 5.0)
• sampling_rate: The sample rate (default: 16000)
• yield: Thai text transcription from each audio chunk

Options for model

• airesearch/wav2vec2-large-xlsr-53-th (default) - AI RESEARCH - PyThaiNLP model
• wannaphong/wav2vec2-large-xlsr-53-th-cv8-newmm - Thai Wav2Vec2 with CommonVoice V8 (newmm tokenizer)
• wannaphong/wav2vec2-large-xlsr-53-th-cv8-deepcut - Thai Wav2Vec2 with CommonVoice V8 (deepcut tokenizer)
• biodatlab/whisper-small-th-combined - Thai Whisper small model
• biodatlab/whisper-th-medium-combined - Thai Whisper medium model
• biodatlab/whisper-th-large-combined - Thai Whisper large model

You can read about models from the list:

• airesearch/wav2vec2-large-xlsr-53-th - AI RESEARCH - PyThaiNLP model
• annaphong/wav2vec2-large-xlsr-53-th-cv8-newmm - Thai Wav2Vec2 with CommonVoice V8 (newmm tokenizer) + language model
• wannaphong/wav2vec2-large-xlsr-53-th-cv8-deepcut - Thai Wav2Vec2 with CommonVoice V8 (deepcut tokenizer) + language model
• biodatlab/whisper-small-th-combined - Thai Whisper small model
• biodatlab/whisper-th-medium-combined - Thai Whisper medium model
• biodatlab/whisper-th-large-combined - Thai Whisper large model

Docker

To use this inside of Docker do the following:

docker build -t <Your Tag name> .
docker run docker run --entrypoint /bin/bash -it <Your Tag name>

You will then get access to a interactive shell environment where you can use python with all packages installed.