README.md

---
language:
- zh
license: apache-2.0
tags:
- whisper-event
- generated_from_trainer
base_model: openai/whisper-small
datasets:
- mozilla-foundation/common_voice_11_0
model-index:
- name: Distil-Whisper Small zh-HK - Alvin
  results:
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: mozilla-foundation/common_voice_16_0 yue
      type: mozilla-foundation/common_voice_16_0
      config: yue
      split: test
      args: yue
    metrics:
    - name: Normalized CER
      type: cer
      value: 9.7
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# Distil-Whisper Small zh-HK - Alvin

- This model is a distilled version of [alvanlii/whisper-small-cantonese](https://huggingface.co/alvanlii/whisper-small-cantonese) on the Cantonese language. 
- Achieves a 9.7 CER (without punctuations), 11.59 CER (with punctuations) on Common Voice 16.0.
- Has 3 decoder layers instead of regular 12 of the Whisper small model.
- Uses ~2GB of GPU VRAM

## Training and evaluation data
For training,
- CantoMap: Winterstein, Grégoire, Tang, Carmen and Lai, Regine (2020) "CantoMap: a Hong Kong Cantonese MapTask Corpus", in Proceedings of The 12th Language Resources and Evaluation Conference, Marseille: European Language Resources Association, p. 2899-2906.
- Cantonse-ASR: Yu, Tiezheng, Frieske, Rita, Xu, Peng, Cahyawijaya, Samuel, Yiu, Cheuk Tung, Lovenia, Holy, Dai, Wenliang, Barezi, Elham, Chen, Qifeng, Ma, Xiaojuan, Shi, Bertram, Fung, Pascale (2022) "Automatic Speech Recognition Datasets in Cantonese: A Survey and New Dataset", 2022. Link: https://arxiv.org/pdf/2201.02419.pdf
- Common Voice yue and zh-HK train sets

For evaluation, Common Voice 16.0 yue Test set is used.

## Comparisons to Whisper Small
||`alvanlii/distil-whisper-small-cantonese`|`alvanlii/whisper-small-cantonese`|
|--|--|--|
|CER (lower is better)|0.097|0.089|
|GPU Inference time (sdpa) [s/sample]|0.027|0.055|
|GPU Inference (regular) [s/sample]|0.027|0.308|
|CPU Inference [s/sample]|1.3|2.57|
|Params [M]|157|242|

Note: inference time is calculated by taking the average inference time for the CV16 yue test set 

## Using the Model
  ```
import librosa

import torch
from transformers import WhisperForConditionalGeneration, WhisperProcessor

y, sr = librosa.load('audio.mp3', sr=16000)

MODEL_NAME = "alvanlii/distil-whisper-small-cantonese"

processor = WhisperProcessor.from_pretrained(MODEL_NAME)
model = WhisperForConditionalGeneration.from_pretrained(MODEL_NAME)

model.config.forced_decoder_ids = None
model.config.suppress_tokens = []
model.config.use_cache = False

processed_in = processor(y, sampling_rate=sr, return_tensors="pt")
gout = model.generate(
    input_features=processed_in.input_features, 
    output_scores=True, return_dict_in_generate=True
)
transcription = processor.batch_decode(gout.sequences, skip_special_tokens=True)[0]
print(transcription)
```
- Alternatively, you can use huggingface pipelines
```
from transformers import pipeline
MODEL_NAME = "alvanlii/distil-whisper-small-cantonese" 
lang = "zh"
pipe = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
)
pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language=lang, task="transcribe")
text = pipe(file)["text"]
```