Model Introduction

Highlight

The model is based on wav2vec2-base and fine-tuned with iemocap and Emotional Speech Dataset (ESD) data, so it supports Chinese and English audio.
The accuracy is as high as 92.9%
The model card shows only part of the source code.See Files and Versions for details
The model can predict the emotion of anger
Some details are as follows

import logging
import pathlib
import re
import sys
import time
import csv
from dataclasses import dataclass, field
from typing import Any, Callable, Dict, List, Optional, Set, Union

import datasets
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from packaging import version
from torch.cuda.amp import GradScaler, autocast

import librosa
from lang_trans import arabic
from datasets import Dataset

import soundfile as sf
from model import Wav2Vec2ForCTCnCLS
from transformers.trainer_utils import get_last_checkpoint

from transformers import (
    HfArgumentParser,
    Trainer,
    TrainingArguments,
    Wav2Vec2CTCTokenizer,
    Wav2Vec2FeatureExtractor,
    Wav2Vec2Processor,
    is_apex_available,
    trainer_utils,
)


local_model_path = "local_model"

if is_apex_available():
    from apex import amp

if version.parse(torch.__version__) >= version.parse("1.6"):
    _is_native_amp_available = True
    from torch.cuda.amp import autocast


logger = logging.getLogger(__name__)

@dataclass
class TrainingArguments(TrainingArguments):
    output_dir: str = field(
        default="output/angry_tmp", metadata={"help": "The store of your output."})
    do_predict: bool = field(
        default=True, metadata={"help": "The store of your output."})
    do_eval: bool = field(
        default=False, metadata={"help": "The store of your output."})
    overwrite_output_dir: str = field(
        default='overwrite_output_dir', metadata={"help": "The store of your output."}   )
    per_device_eval_batch_size: int = field(
        default=2, metadata={"help": "The store of your output."})
    warmup_ratio: float = field(
        default=0.1, metadata={"help": "Linear warmup over warmup_ratio fraction of total steps."}
    )
   


@dataclass
class DataCollatorCTCWithPadding:
    """
    Data collator that will dynamically pad the inputs received.
    Args:
        processor (:class:`~transformers.Wav2Vec2Processor`)
            The processor used for proccessing the data.
        padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):
            Select a strategy to pad the returned sequences (according to the model's padding side and padding index)
            among:
            * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
              sequence if provided).
            * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the
              maximum acceptable input length for the model if that argument is not provided.
            * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of
              different lengths).
        max_length (:obj:`int`, `optional`):
            Maximum length of the ``input_values`` of the returned list and optionally padding length (see above).
        max_length_labels (:obj:`int`, `optional`):
            Maximum length of the ``labels`` returned list and optionally padding length (see above).
        pad_to_multiple_of (:obj:`int`, `optional`):
            If set will pad the sequence to a multiple of the provided value.
            This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability >=
            7.5 (Volta).
    """

    processor: Wav2Vec2Processor
    padding: Union[bool, str] = True
    max_length: Optional[int] = None
    max_length_labels: Optional[int] = None
    pad_to_multiple_of: Optional[int] = None
    pad_to_multiple_of_labels: Optional[int] = None
    audio_only = False
    duration = 6
    sample_rate = 16000
    

    def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:
        # split inputs and labels since they have to be of different lenghts and need
        # different padding methods
        input_features = [{"input_values": feature["input_values"]} for feature in features]
        
        batch = self.processor.pad(
            input_features,
            padding=self.padding,
            # max_length=self.max_length,
            max_length=self.duration*self.sample_rate,
            pad_to_multiple_of=self.pad_to_multiple_of,
            return_tensors="pt",
        )

        return batch


class CTCTrainer(Trainer):
    def _prepare_inputs(self, inputs: Dict[str, Union[torch.Tensor, Any]]) -> Dict[str, Union[torch.Tensor, Any]]:
        self.use_amp = False
        self.use_apex = False
        self.deepspeed = False
        self.scaler = GradScaler() 
        for k, v in inputs.items():
            if isinstance(v, torch.Tensor):
                kwargs = dict(device=self.args.device)
                if self.deepspeed and inputs[k].dtype != torch.int64:
                    kwargs.update(dict(dtype=self.args.hf_deepspeed_config.dtype()))
                inputs[k] = v.to(**kwargs)

        if self.args.past_index >= 0 and self._past is not None:
            inputs["mems"] = self._past

        return inputs

    
def create_dataset(audio_path):
    data = {
        'file': [audio_path]
    }
    dataset = Dataset.from_dict(data)
    return dataset


def exeute_angry_predict(audio_path):
    # See all possible arguments in src/transformers/training_args.py
    # or by passing the --help flag to this script.
    # We now keep distinct sets of args, for a cleaner separation of concerns.
    
    target_sr = 16000

    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))

    model_args, data_args, training_args = parser.parse_args_into_dataclasses()
    configure_logger(model_args, training_args)

   
    orthography = Orthography.from_name(data_args.orthography.lower())
    orthography.tokenizer = model_args.tokenizer
    processor = orthography.create_processor(model_args)

    if data_args.dataset_name == 'emotion':
        val_dataset = create_dataset(audio_path)
        cls_label_map = {"neutral":0, "angry":1}

    model = Wav2Vec2ForCTCnCLS.from_pretrained(
        local_model_path,
        gradient_checkpointing=True, # training_args.gradient_checkpointing,
        cls_len=len(cls_label_map),
    )

    def prepare_example(example, audio_only=False):  # TODO(elgeish) make use of multiprocessing?
        example["speech"], example["sampling_rate"] = librosa.load(example[data_args.speech_file_column], sr=target_sr)
        orig_sample_rate = example["sampling_rate"]
        target_sample_rate = target_sr
        if orig_sample_rate != target_sample_rate:
            example["speech"] = librosa.resample(example["speech"], orig_sr=orig_sample_rate, target_sr=target_sample_rate)
        if data_args.max_duration_in_seconds is not None:
            example["duration_in_seconds"] = len(example["speech"]) / example["sampling_rate"]
        return example


    if training_args.do_predict:
        val_dataset = val_dataset.map(prepare_example, fn_kwargs={'audio_only':True})
    
   
    def prepare_dataset(batch, audio_only=False):
        # check that all files have the correct sampling rate
        assert (
            len(set(batch["sampling_rate"])) == 1
        ), f"Make sure all inputs have the same sampling rate of {processor.feature_extractor.sampling_rate}."
        
        batch["input_values"] = processor(batch["speech"], sampling_rate=batch["sampling_rate"][0]).input_values
        return batch

    
    if training_args.do_predict:
        val_dataset = val_dataset.map(
            prepare_dataset,
            fn_kwargs={'audio_only':True},
            batch_size=training_args.per_device_eval_batch_size,
            batched=True,
            num_proc=data_args.preprocessing_num_workers,
        )

    data_collator = DataCollatorCTCWithPadding(processor=processor, padding=True)

    if model_args.freeze_feature_extractor:
        model.freeze_feature_extractor()

    trainer = CTCTrainer(
        model=model,
        args=training_args,
        eval_dataset=val_dataset,
        tokenizer=processor.feature_extractor,
    )


    if training_args.do_predict:
        logger.info('******* Predict ********')
        data_collator.audio_only=True
        results= {}
        result= ''
        predictions, labels, metrics = trainer.predict(val_dataset, metric_key_prefix="predict")
        logits_ctc, logits_cls = predictions
        pred_ids = np.argmax(logits_cls, axis=-1)
        if pred_ids==0:
            result = "neutral"
        if pred_ids==1:
            result = "angry"
        results[audio_path] = result
        print("results", results)
        

if __name__ == "__main__":
    audio_path = 'audio.mp3'
    exeute_angry_predict(audio_path)