metadata

language:
  - kk
tags:
  - audio
  - automatic-speech-recognition
  - kazakh-asr
widget:
  - src: >-
      https://drive.google.com/file/d/1udN8ybS7Ih3ESuoYZlaei4RcIPVbJlAf/view?usp=sharing
    example_title: sample
model-index:
  - name: whisper-base.kk
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Kazakh Speech Corpus 2 (KSC2)
          type: librispeech_asr
          config: clean
          split: test
          args:
            language: kk
        metrics:
          - name: Test WER
            type: wer
            value: 15.36
pipeline_tag: automatic-speech-recognition
license: apache-2.0

Whisper

Whisper-base for automatic speech recognition (ASR) for the low-resourced Kazakh language. The model was fine-tuned on the Kazakh Speech Corpus 2 with over 1k hours of labelled data. The model achieved 15.36% WER on the test set.

Usage

This checkpoint is a Kazakh-only model, meaning it can be used only for Kazakh speech recognition.

Transcription

>>> from transformers import WhisperProcessor, WhisperForConditionalGeneration
>>> import librosa

>>> # load model and processor
>>> processor = WhisperProcessor.from_pretrained("akuzdeuov/whisper-base.kk")
>>> model = WhisperForConditionalGeneration.from_pretrained("akuzdeuov/whisper-base.kk")

>>> # load your audio
>>> audio, sampling_rate = librosa.load("path_to_audio", sr=16000)
>>> input_features = processor(audio, sampling_rate=sampling_rate, return_tensors="pt").input_features 

>>> # generate token ids
>>> predicted_ids = model.generate(input_features)
>>> # decode token ids to text
>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=False)

>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)

The context tokens can be removed from the start of the transcription by setting skip_special_tokens=True.

Long-Form Transcription

The Whisper model is intrinsically designed to work on audio samples of up to 30s in duration. However, by using a chunking algorithm, it can be used to transcribe audio samples of up to arbitrary length. This is possible through Transformers pipeline method. Chunking is enabled by setting chunk_length_s=30 when instantiating the pipeline. With chunking enabled, the pipeline can be run with batched inference.

>>> import torch
>>> from transformers import pipeline

>>> device = "cuda:0" if torch.cuda.is_available() else "cpu"

>>> pipe = pipeline(
>>>   "automatic-speech-recognition",
>>>   model="akuzdeuov/whisper-base.kk",
>>>   chunk_length_s=30,
>>>   device=device,
>>> )

>>> prediction = pipe("path_to_audio", batch_size=8)["text"]

References

Whisper, OpenAI.