Skip to content

Latest commit

 

History

History
135 lines (86 loc) · 3.52 KB

speech-transcription-vosk.md

File metadata and controls

135 lines (86 loc) · 3.52 KB
Raw

vosk module

The vosk module contains the VoskLearner class, which inherits from the abstract class Learner.

Class VoskLearner

Bases: engine.learners.Learner

The VoskLearner class is a wrapper of libary [1] implementation. It is integrated for the speech transcription task.

The VoskLearner class has the following public methods:

VoskLearner constructor

VoskLearner(self, device, sample_rate)

Constructor parameters:

  • device: str, default="cpu"
    The device to use for computations. Currently only supports cpu.

  • sample_rate: int, default=16000
    The sample rate to be used by the Vosk model.

VoskLearner.eval

VoskLearner.eval(self, dataset, save_path_csv)

This method is used to evaluate Vosk model on the given dataset.

Returns a dictionary containing evaluation metrics such as word error rate.

Parameters:

  • dataset: DatasetIterator
    A speech dataset.
  • save_path_csv: Optional[str], default=None
    The path to save the evaluation results.

VoskLearner.infer

VoskLearner.infer(self, audio)

This method runs inference on an audio sample. Please call the load() method before calling this method.

Return transcription as VoskTranscription that contains transcription text and other side information.

Parameters:

  • audio: Union[Timeseries, torch.Tensor, np.ndarray, bytes]
    The audio sample as a Timeseries, torch.Tensor, or np.ndarray or bytes.

VoskLearner.load

VoskLearner.load(self, name, language, model_path, download_dir)

This method loads the Vosk model and initializes the recognizer. The method will download model if necessary

Parameters:

  • name: Optional[str], default=None
    Full name of the Vosk model.

  • language: Optional[str], default=None
    Language of the Vosk model. Vosk will decide the default model for this language.

  • model_path: Optional[str], default=None
    Path to the Vosk model.

  • download_dir: Optional[bool], default=False
    Directory to download the Vosk model to.

VoskLearner.download

VoskLearner.download(self, model_name)

Download model given a local path including the full name of the Vosk model.

Parameters:

  • model_name: Path
    Path to download model to, including the full name of the Vosk model.

VoskLearner.reset

VoskLearner.reset(self)

This method sets Vosk model, model name, language, and KalidRecognizer to None. Use before load a new model.

VoskLearner.reset_rec

VoskLearner.reset_rec(self)

This method resets the KalidRecognizer.

Examples

  • Download and load a model by its name and infer a sample from an existing file.
import librosa
import numpy as np

from opendr.engine.data import Timeseries
from opendr.perception.speech_transcription import VoskLearner

learner = VoskLearner()
learner.load(language="en-us")

# Assuming you have recorded your own voice sample in command.wav in the current directory
signal, sampling_rate = librosa.load("video.wav", sr=learner.sample_rate)
signal = np.expand_dims(signal, axis=0)
timeseries = Timeseries(signal)
result = learner.infer(timeseries)
print(result)

References

[1] Github: alphacep/vosk-api.