Whisper Large Zh Cv11

🚀 Whisper Large Chinese (Mandarin)

This model is a fine - tuned version of openai/whisper-large-v2 on Chinese (Mandarin). It uses the train and validation splits of Common Voice 11. Note that not all validation split data were used during training. 1k samples were extracted from the validation split for evaluation during fine - tuning.

🚀 Quick Start

Installation

To use this model, you need to install the transformers library. You can install it using the following command:

pip install transformers

💻 Usage Examples

Basic Usage

from transformers import pipeline

transcriber = pipeline(
  "automatic-speech-recognition", 
  model="jonatasgrosman/whisper-large-zh-cv11"
)

transcriber.model.config.forced_decoder_ids = (
  transcriber.tokenizer.get_decoder_prompt_ids(
    language="zh", 
    task="transcribe"
  )
)

transcription = transcriber("path/to/my_audio.wav")

📚 Documentation

Evaluation

The model was evaluated using the test split of two datasets: Common Voice 11 (the same dataset used for fine - tuning) and Fleurs (a dataset not seen during fine - tuning).

Since Whisper can transcribe casing and punctuation, the model evaluation was performed in two different scenarios: one using the raw text and the other using the normalized text (lowercase + removal of punctuations). Additionally, for the Fleurs dataset, the model was evaluated in a scenario where only non - numeric samples were kept because the way numerical values are described in this dataset is different from that in the fine - tuning dataset (Common Voice), which may affect the model's performance for this type of transcription.

Common Voice 11

Fleurs

📄 License

This project is licensed under the Apache - 2.0 license.

📋 Model Information