---
license: apache-2.0
base_model: openai/whisper-large-v3
tags:
- generated_from_trainer
metrics:
- wer
model-index:
- name: Hibiki_ASR_Phonemizer
  results: []
language:
- ja
---

# Hibiki ASR Phonemizer

This model is a Phoneme Level Speech Recognition network, originally a fine-tuned version of [openai/whisper-large-v3](https://huggingface.co/openai/whisper-large-v3) on a
mixture of Different Japanese datasets.

it can detect, transcribe and do the following:

- non-speech sounds such as gasp, erotic moans, laughter, etc.
- adding punctuations more faithfully.

a Grapheme decoder head (i.e outputting normal Japanese) will probably be trained as well. Though going directly from audio to Phonemes will result in a
more accurate representation for Japanese.



## Inference and Post-proc (Highly recommended to check the notebook below!)

```python

# this function was borrowed and modified from Aaron Yinghao Li, the Author of StyleTTS paper.

from datasets import Dataset, Audio
from transformers import WhisperProcessor, WhisperForConditionalGeneration
import re
import pykakasi

kana_mapper = dict([
    ("ゔぁ","ba"),
          .
          .
          .
          etc. # Take a look at the Notebook for the whole code
    ("ぉ"," o"),
    ("ゎ"," ɯa"),
    ("ぉ"," o"),

    ("を","o")
])


def post_fix(text):
    orig = text

    for k, v in kana_mapper.items():
        text = text.replace(k, v)

    return text


processor = WhisperProcessor.from_pretrained("openai/whisper-large-v3")
model = WhisperForConditionalGeneration.from_pretrained("Respair/Hibiki_ASR_Phonemizer").to("cuda:0")

forced_decoder_ids = processor.get_decoder_prompt_ids(task="transcribe", language='japanese')




def convert_to_kana(text):
    kks = pykakasi.kakasi()


    def convert_word(word):
        result = kks.convert(word)
        return ''.join(item['hira'] for item in result)


    parts = re.split(r'([^\u3000-\u30ff\u3400-\u4dbf\u4e00-\u9fff]+)', text)


    converted_parts = [convert_word(part) if re.match(r'[\u3000-\u30ff\u3400-\u4dbf\u4e00-\u9fff]', part) else part for part in parts]

    return ''.join(converted_parts)


sample = Dataset.from_dict({"audio": ["/content/kl_chunk1987.wav"]}).cast_column("audio", Audio(16000))
sample = sample[0]['audio']

# Ensure the input features are on the same device as the model
input_features = processor(sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt").input_features.to("cuda:0")

# generate token ids
predicted_ids = model.generate(input_features,forced_decoder_ids=forced_decoder_ids, repetition_penalty=1.2)
# decode token ids to text
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)


# You can add your final adjustments here, it's better to write a dict though, but I'm just giving you a quick demonstration here.

if ' neɽitai ' in transcription[0]:
    transcription[0] = transcription[0].replace(' neɽitai ', "naɽitai")

if 'harɯdʑisama' in transcription[0]:
    transcription[0] = transcription[0].replace('harɯdʑisama', "arɯdʑisama")


if 'tɕabiʔto' in transcription[0]:
    transcription[0] = transcription[0].replace('tɕabiʔto', "tɕabiʔto")


if "ki ni ɕinai" in transcription[0]:
    transcription[0] = re.sub(r'(?<!\s)ki ni ɕinai', r' ki ni ɕinai', transcription[0])

if 'ʔt' in transcription[0]:
    transcription[0] = re.sub(r'(?<!\s)ʔt', r'ʔt', transcription[0])

if 'de aɽoɯ' in transcription[0]:
    transcription[0] = re.sub(r'(?<!\s)de aɽoɯ', r' de aɽoɯ', transcription[0])

if ".ʔ" in transcription[0]:
    transcription[0] = transcription[0].replace(".ʔ","..")

if "ʔ." in transcription[0]:
    transcription[0] = transcription[0].replace("ʔ.",".")

transcription[0] = convert_to_kana(transcription[0]) # Ensuring the model won't hallucinate and accidentally return kana / kanji.

post_fix(transcription[0].lstrip())

```

the Full code -> [Notebook](https://colab.research.google.com/drive/13tx8WKzkvePFdtKU4WUE_iYyYCqTY8dZ#scrollTo=5XqUs-sPdT79)

## Intended uses & limitations

No restrictions is imposed by me, but proceed at your own risk, The User (You) are entirely responisble for their actions.

## Training and evaluation data

- Japanese Common Voice 17
- ehehe Corpus
- Custom Game and Anime dataset (around 8 hours)

## Training procedure

### Training hyperparameters


The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 24
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 5000

### Compute and Duration

- 1x A100(40G)
- 64gb RAM
- BF16
- 14hrs

### Framework versions

- Transformers 4.41.1
- Pytorch 2.4.0+cu121
- Datasets 2.19.1
- Tokenizers 0.19.1