timit.py

# coding=utf-8

"""TIMIT automatic speech recognition and speaker identification dataset."""


import os
import textwrap
import datasets
import itertools
import typing as tp
from pathlib import Path


SAMPLE_RATE = 16_000

_ZIP_FILENAME = 'timit.tar.gz'

_CITATION = """\
@inproceedings{
  title={TIMIT Acoustic-Phonetic Continuous Speech Corpus},
  author={Garofolo, John S., et al},
  ldc_catalog_no={LDC93S1},
  DOI={https://doi.org/10.35111/17gk-bn40},
  journal={Linguistic Data Consortium, Philadelphia},
  year={1983}
}
"""

_DESCRIPTION = """\
The TIMIT corpus of reading speech has been developed to provide speech data for acoustic-phonetic research studies
and for the evaluation of automatic speech recognition systems.
TIMIT contains high quality recordings of 630 individuals/speakers with 8 different American English dialects,
with each individual reading upto 10 phonetically rich sentences.
More info on TIMIT dataset can be understood from the "README" which can be found here:
https://catalog.ldc.upenn.edu/docs/LDC93S1/readme.txt
"""

_HOMEPAGE = "https://catalog.ldc.upenn.edu/LDC93S1"
_SPEAKERS = ['FADG0', 'FAEM0', 'FAJW0', 'FAKS0', 'FALK0', 'FALR0', 'FAPB0', 'FASW0', 'FAWF0', 'FBAS0', 'FBCG1', 'FBCH0', 'FBJL0', 'FBLV0', 'FBMH0', 'FBMJ0', 'FCAG0', 'FCAJ0', 'FCAL1', 'FCAU0', 'FCDR1', 'FCEG0', 'FCFT0', 'FCJF0', 'FCJS0', 'FCKE0', 'FCLT0', 'FCMG0', 'FCMH0', 'FCMH1', 'FCMM0', 'FCMR0', 'FCRH0', 'FCRZ0', 'FCYL0', 'FDAC1', 'FDAS1', 'FDAW0', 'FDFB0', 'FDHC0', 'FDJH0', 'FDKN0', 'FDML0', 'FDMS0', 'FDMY0', 'FDNC0', 'FDRD1', 'FDRW0', 'FDTD0', 'FDXW0', 'FEAC0', 'FEAR0', 'FECD0', 'FEDW0', 'FEEH0', 'FELC0', 'FEME0', 'FETB0', 'FEXM0', 'FGCS0', 'FGDP0', 'FGJD0', 'FGMB0', 'FGMD0', 'FGRW0', 'FGWR0', 'FHES0', 'FHEW0', 'FHLM0', 'FHXS0', 'FISB0', 'FJAS0', 'FJCS0', 'FJDM2', 'FJEM0', 'FJEN0', 'FJHK0', 'FJKL0', 'FJLG0', 'FJLM0', 'FJLR0', 'FJMG0', 'FJRB0', 'FJRE0', 'FJRP1', 'FJSA0', 'FJSJ0', 'FJSK0', 'FJSP0', 'FJWB0', 'FJWB1', 'FJXM0', 'FJXP0', 'FKAA0', 'FKDE0', 'FKDW0', 'FKFB0', 'FKKH0', 'FKLC0', 'FKLC1', 'FKLH0', 'FKMS0', 'FKSR0', 'FLAC0', 'FLAG0', 'FLAS0', 'FLBW0', 'FLEH0', 'FLET0', 'FLHD0', 'FLJA0', 'FLJD0', 'FLJG0', 'FLKD0', 'FLKM0', 'FLMA0', 'FLMC0', 'FLMK0', 'FLNH0', 'FLOD0', 'FLTM0', 'FMAF0', 'FMAH0', 'FMAH1', 'FMBG0', 'FMCM0', 'FMEM0', 'FMGD0', 'FMJB0', 'FMJF0', 'FMJU0', 'FMKC0', 'FMKF0', 'FMLD0', 'FMMH0', 'FMML0', 'FMPG0', 'FNKL0', 'FNLP0', 'FNMR0', 'FNTB0', 'FPAB1', 'FPAC0', 'FPAD0', 'FPAF0', 'FPAS0', 'FPAZ0', 'FPJF0', 'FPKT0', 'FPLS0', 'FPMY0', 'FRAM1', 'FREH0', 'FREW0', 'FRJB0', 'FRLL0', 'FRNG0', 'FSAG0', 'FSAH0', 'FSAK0', 'FSBK0', 'FSCN0', 'FSDC0', 'FSDJ0', 'FSEM0', 'FSGF0', 'FSJG0', 'FSJK1', 'FSJS0', 'FSJW0', 'FSKC0', 'FSKL0', 'FSKP0', 'FSLB1', 'FSLS0', 'FSMA0', 'FSMM0', 'FSMS1', 'FSPM0', 'FSRH0', 'FSSB0', 'FSXA0', 'FTAJ0', 'FTBR0', 'FTBW0', 'FTLG0', 'FTLH0', 'FTMG0', 'FUTB0', 'FVFB0', 'FVKB0', 'FVMH0', 'MABC0', 'MABW0', 'MADC0', 'MADD0', 'MAEB0', 'MAEO0', 'MAFM0', 'MAHH0', 'MAJC0', 'MAJP0', 'MAKB0', 'MAKR0', 'MAPV0', 'MARC0', 'MARW0', 'MBAR0', 'MBBR0', 'MBCG0', 'MBDG0', 'MBEF0', 'MBGT0', 'MBJK0', 'MBJV0', 'MBMA0', 'MBMA1', 'MBML0', 'MBNS0', 'MBOM0', 'MBPM0', 'MBSB0', 'MBTH0', 'MBWM0', 'MBWP0', 'MCAE0', 'MCAL0', 'MCCS0', 'MCDC0', 'MCDD0', 'MCDR0', 'MCEF0', 'MCEM0', 'MCEW0', 'MCHH0', 'MCHL0', 'MCLK0', 'MCLM0', 'MCMB0', 'MCMJ0', 'MCPM0', 'MCRC0', 'MCRE0', 'MCSH0', 'MCSS0', 'MCTH0', 'MCTM0', 'MCTT0', 'MCTW0', 'MCXM0', 'MDAB0', 'MDAC0', 'MDAC2', 'MDAS0', 'MDAW1', 'MDBB0', 'MDBB1', 'MDBP0', 'MDCD0', 'MDCM0', 'MDDC0', 'MDED0', 'MDEF0', 'MDEM0', 'MDHL0', 'MDHS0', 'MDJM0', 'MDKS0', 'MDLB0', 'MDLC0', 'MDLC1', 'MDLC2', 'MDLD0', 'MDLF0', 'MDLH0', 'MDLM0', 'MDLR0', 'MDLR1', 'MDLS0', 'MDMA0', 'MDMT0', 'MDNS0', 'MDPB0', 'MDPK0', 'MDPS0', 'MDRB0', 'MDRD0', 'MDRM0', 'MDSC0', 'MDSJ0', 'MDSS0', 'MDSS1', 'MDTB0', 'MDVC0', 'MDWA0', 'MDWD0', 'MDWH0', 'MDWK0', 'MDWM0', 'MEAL0', 'MEDR0', 'MEFG0', 'MEGJ0', 'MEJL0', 'MEJS0', 'MERS0', 'MESD0', 'MESG0', 'MESJ0', 'MEWM0', 'MFER0', 'MFGK0', 'MFMC0', 'MFRM0', 'MFWK0', 'MFXS0', 'MFXV0', 'MGAF0', 'MGAG0', 'MGAK0', 'MGAR0', 'MGAW0', 'MGES0', 'MGJC0', 'MGJF0', 'MGLB0', 'MGMM0', 'MGRL0', 'MGRP0', 'MGRT0', 'MGSH0', 'MGSL0', 'MGWT0', 'MGXP0', 'MHBS0', 'MHIT0', 'MHJB0', 'MHMG0', 'MHMR0', 'MHPG0', 'MHRM0', 'MHXL0', 'MILB0', 'MJAC0', 'MJAE0', 'MJAI0', 'MJAR0', 'MJBG0', 'MJBR0', 'MJDA0', 'MJDC0', 'MJDE0', 'MJDG0', 'MJDH0', 'MJDM0', 'MJDM1', 'MJEB0', 'MJEB1', 'MJEE0', 'MJES0', 'MJFC0', 'MJFH0', 'MJFR0', 'MJHI0', 'MJJB0', 'MJJG0', 'MJJJ0', 'MJJM0', 'MJKR0', 'MJLB0', 'MJLG1', 'MJLN0', 'MJLS0', 'MJMA0', 'MJMD0', 'MJMM0', 'MJMP0', 'MJPG0', 'MJPM0', 'MJPM1', 'MJRA0', 'MJRF0', 'MJRG0', 'MJRH0', 'MJRH1', 'MJRK0', 'MJRP0', 'MJSR0', 'MJSW0', 'MJTC0', 'MJTH0', 'MJVW0', 'MJWG0', 'MJWS0', 'MJWT0', 'MJXA0', 'MJXL0', 'MKAG0', 'MKAH0', 'MKAJ0', 'MKAM0', 'MKCH0', 'MKCL0', 'MKDB0', 'MKDD0', 'MKDR0', 'MKDT0', 'MKES0', 'MKJL0', 'MKJO0', 'MKLN0', 'MKLR0', 'MKLS0', 'MKLS1', 'MKLT0', 'MKLW0', 'MKRG0', 'MKXL0', 'MLBC0', 'MLEL0', 'MLIH0', 'MLJB0', 'MLJC0', 'MLJH0', 'MLLL0', 'MLNS0', 'MLNT0', 'MLSH0', 'MMAA0', 'MMAB0', 'MMAB1', 'MMAG0', 'MMAM0', 'MMAR0', 'MMBS0', 'MMCC0', 'MMDB0', 'MMDB1', 'MMDG0', 'MMDH0', 'MMDM0', 'MMDM1', 'MMDM2', 'MMDS0', 'MMEA0', 'MMEB0', 'MMGC0', 'MMGG0', 'MMGK0', 'MMJB1', 'MMJR0', 'MMLM0', 'MMPM0', 'MMRP0', 'MMSM0', 'MMVP0', 'MMWB0', 'MMWH0', 'MMWS0', 'MMWS1', 'MMXS0', 'MNET0', 'MNJM0', 'MNLS0', 'MNTW0', 'MPAB0', 'MPAM0', 'MPAM1', 'MPAR0', 'MPCS0', 'MPDF0', 'MPEB0', 'MPFU0', 'MPGH0', 'MPGL0', 'MPGR0', 'MPGR1', 'MPLB0', 'MPMB0', 'MPPC0', 'MPRB0', 'MPRD0', 'MPRK0', 'MPRT0', 'MPSW0', 'MPWM0', 'MRAB0', 'MRAB1', 'MRAI0', 'MRAM0', 'MRAV0', 'MRBC0', 'MRCG0', 'MRCS0', 'MRCW0', 'MRCZ0', 'MRDD0', 'MRDM0', 'MRDS0', 'MREB0', 'MREE0', 'MREH1', 'MREM0', 'MRES0', 'MREW1', 'MRFK0', 'MRFL0', 'MRGG0', 'MRGM0', 'MRGS0', 'MRHL0', 'MRJB1', 'MRJH0', 'MRJM0', 'MRJM1', 'MRJM3', 'MRJM4', 'MRJO0', 'MRJR0', 'MRJS0', 'MRJT0', 'MRKM0', 'MRKO0', 'MRLD0', 'MRLJ0', 'MRLJ1', 'MRLK0', 'MRLR0', 'MRMB0', 'MRMG0', 'MRMH0', 'MRML0', 'MRMS0', 'MRMS1', 'MROA0', 'MRPC0', 'MRPC1', 'MRPP0', 'MRRE0', 'MRRK0', 'MRSO0', 'MRSP0', 'MRTC0', 'MRTJ0', 'MRTK0', 'MRVG0', 'MRWA0', 'MRWS0', 'MRWS1', 'MRXB0', 'MSAH1', 'MSAS0', 'MSAT0', 'MSAT1', 'MSDB0', 'MSDH0', 'MSDS0', 'MSEM1', 'MSES0', 'MSFH0', 'MSFH1', 'MSFV0', 'MSJK0', 'MSJS1', 'MSLB0', 'MSMC0', 'MSMR0', 'MSMS0', 'MSRG0', 'MSRR0', 'MSTF0', 'MSTK0', 'MSVS0', 'MTAA0', 'MTAB0', 'MTAS0', 'MTAS1', 'MTAT0', 'MTAT1', 'MTBC0', 'MTCS0', 'MTDB0', 'MTDP0', 'MTDT0', 'MTEB0', 'MTER0', 'MTHC0', 'MTJG0', 'MTJM0', 'MTJS0', 'MTJU0', 'MTKD0', 'MTKP0', 'MTLB0', 'MTLC0', 'MTLS0', 'MTML0', 'MTMN0', 'MTMR0', 'MTMT0', 'MTPF0', 'MTPG0', 'MTPP0', 'MTPR0', 'MTQC0', 'MTRC0', 'MTRR0', 'MTRT0', 'MTWH0', 'MTWH1', 'MTXS0', 'MVJH0', 'MVLO0', 'MVRW0', 'MWAC0', 'MWAD0', 'MWAR0', 'MWBT0', 'MWCH0', 'MWDK0', 'MWEM0', 'MWEW0', 'MWGR0', 'MWJG0', 'MWRE0', 'MWRP0', 'MWSB0', 'MWSH0', 'MWVW0', 'MZMB0']


class TimitConfig(datasets.BuilderConfig):
    """BuilderConfig for TIMIT."""
    
    def __init__(self, features, **kwargs):
        super(TimitConfig, self).__init__(version=datasets.Version("0.0.1", ""), **kwargs)
        self.features = features


class TIMIT(datasets.GeneratorBasedBuilder):

    BUILDER_CONFIGS = [
        TimitConfig(
            features=datasets.Features(
                {
                    "file": datasets.Value("string"),
                    "audio": datasets.Audio(sampling_rate=SAMPLE_RATE),
                    "speaker_id": datasets.Value("string"),
                    "label": datasets.ClassLabel(names=_SPEAKERS),
                }
            ),
            name="si", 
            description=textwrap.dedent(
                """\
                Speaker Identification (SI) classifies each utterance for its speaker identity as a multi-class
                classification, where speakers are in the same pre-defined set for both training and testing. 
                The evaluation metric is accuracy (ACC).
                """
            ),
        ), 
        TimitConfig(
            features=datasets.Features(
                {
                    "file": datasets.Value("string"),
                    "audio": datasets.Audio(sampling_rate=SAMPLE_RATE), 
                    "phonetic_detail": datasets.Sequence(
                        {
                            "start": datasets.Value("int64"),
                            "stop": datasets.Value("int64"),
                            "utterance": datasets.Value("string"),
                        }
                    ),
                    "word_detail": datasets.Sequence(
                        {
                            "start": datasets.Value("int64"),
                            "stop": datasets.Value("int64"),
                            "utterance": datasets.Value("string"),
                        }
                    ),
                    "text": datasets.Value("string"),
                }
            ),
            name="asr", 
            description=textwrap.dedent(
                """\
                ASR transcribes utterances into words. While PR analyses the
                improvement in modeling phonetics, ASR reflects the significance of
                the improvement in a real-world scenario. 
                The evaluation metric is word error rate (WER).
                """
            ),
        )
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=self.config.features,
            supervised_keys=None,
            homepage=_HOMEPAGE,
            citation=_CITATION,
            task_templates=None,
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        archive_path = dl_manager.extract(_ZIP_FILENAME)

        if self.config.name == "si":
            return [
                datasets.SplitGenerator(
                    name=datasets.Split.TRAIN, gen_kwargs={"archive_path": archive_path, "split": "train"}
                ), 
                datasets.SplitGenerator(
                    name=datasets.Split.VALIDATION, gen_kwargs={"archive_path": archive_path, "split": "validation"}
                ), 
                datasets.SplitGenerator(
                    name=datasets.Split.TEST, gen_kwargs={"archive_path": archive_path, "split": "test"}
                ), 
            ]
        elif self.config.name == 'asr':
            return [
                datasets.SplitGenerator(
                    name=datasets.Split.TRAIN, gen_kwargs={"archive_path": archive_path, "split": "train"}
                ), 
                datasets.SplitGenerator(
                    name=datasets.Split.TEST, gen_kwargs={"archive_path": archive_path, "split": "test"}
                ), 
            ]

    def _generate_examples(self, archive_path, split=None):
        if self.config.name == 'si':
            extensions = ['.wav']
            _, _walker = fast_scandir(archive_path, extensions, recursive=True)
            _train_walker, _val_walker, _test_walker = [], [], []
            _pair_list = [(Path(fileid).parent.stem, fileid) for fileid in _walker]
            for key, group in itertools.groupby(_pair_list, lambda x: x[0]):
                _files_per_speaker = [item[1] for item in group]
                _train_walker.extend(_files_per_speaker[:6])
                _val_walker.extend(_files_per_speaker[6:8])
                _test_walker.extend(_files_per_speaker[8:])
            if split == 'train':
                files = _train_walker
            elif split == 'validation':
                files = _val_walker
            elif split == 'test':
                files = _test_walker
            for guid, audio_path in enumerate(files):
                yield guid, {
                    "id": str(guid),
                    "file": audio_path, 
                    "audio": audio_path, 
                    "speaker_id": Path(audio_path).parent.stem, 
                    "label": Path(audio_path).parent.stem, 
                }
        elif self.config.name == 'asr':
            wav_paths = sorted(Path(archive_path).glob(f"**/{split}/**/*.wav"))
            wav_paths = wav_paths if wav_paths else sorted(Path(archive_path).glob(f"**/{split.upper()}/**/*.wav"))
            for guid, wav_path in enumerate(wav_paths):
                # Extract phonemes
                phn_path = with_case_insensitive_suffix(Path(str(wav_path).replace('.wav', '')), ".phn")
                with phn_path.open(encoding="utf-8") as op:
                    phonetic_detail = [
                        {
                            "start": i.split(" ")[0],
                            "stop": i.split(" ")[1],
                            "utterance": " ".join(i.split(" ")[2:]).strip(),
                        }
                        for i in op.readlines()
                    ]
    
                # Extract words
                txt_path = with_case_insensitive_suffix(Path(str(wav_path).replace('.wav', '')), ".wrd")
                with txt_path.open(encoding="utf-8") as op:
                    word_detail = [
                        {
                            "start": i.split(" ")[0],
                            "stop": i.split(" ")[1],
                            "utterance": " ".join(i.split(" ")[2:]).strip(),
                        }
                        for i in op.readlines()
                    ]
                    text = ' '.join([w['utterance'] for w in word_detail])
                yield guid, {
                    "id": str(guid),
                    "file": str(wav_path), 
                    "audio": str(wav_path), 
                    "phonetic_detail": phonetic_detail, 
                    "word_detail": word_detail, 
                    "text": text, 
                }


def fast_scandir(path: str, exts: tp.List[str], recursive: bool = False):
    # Scan files recursively faster than glob
    # From github.com/drscotthawley/aeiou/blob/main/aeiou/core.py
    subfolders, files = [], []

    try:  # hope to avoid 'permission denied' by this try
        for f in os.scandir(path):
            try:  # 'hope to avoid too many levels of symbolic links' error
                if f.is_dir():
                    subfolders.append(f.path)
                elif f.is_file():
                    if os.path.splitext(f.name)[1].lower() in exts:
                        files.append(f.path)
            except Exception:
                pass
    except Exception:
        pass

    if recursive:
        for path in list(subfolders):
            sf, f = fast_scandir(path, exts, recursive=recursive)
            subfolders.extend(sf)
            files.extend(f)  # type: ignore

    return subfolders, files


def with_case_insensitive_suffix(path: Path, suffix: str):
    path = path.with_suffix(suffix.lower())
    path = path if path.exists() else path.with_suffix(suffix.upper())
    return path