RomanianSpeechSynthesis.py

#RECORDING INFORMATION
#
#The Romanian speech synthesis (RSS) corpus was recorded in a hemianechoic chamber (anechoic walls and ceiling; floor partially anechoic) at the University of Edinburgh. We used three high quality studio microphones: a Neumann u89i (large diaphragm condenser), a Sennheiser MKH 800 (small diaphragm condenser with very wide bandwidth) and a DPA 4035 (headset-mounted condenser). Although the current release includes only speech data recorded via Sennheiser MKH 800, we may release speech data recorded via other microphones in the future. All recordings were made at 96 kHz sampling frequency and 24 bits per sample, then downsampled to 48 kHz sampling frequency. For recording, downsampling and bit rate conversion, we used ProTools HD hardware and software. We conducted 8 sessions over the course of a month, recording about 500 sentences in each session. At the start of each session, the speaker listened to a previously recorded sample, in order to attain a similar voice quality and intonation.
#
#ACKNOWLEDGEMENTS
#
#The RSS database is constructed by:
#
#       Adriana Stan      (Technical University of Cluj-Napoca, Romania)
#       Junichi Yamagishi (University of Edinburgh, United Kingdom)
#       Simon King        (University of Edinburgh, United Kingdom)
#       Matthew Aylett    (Cereproc)
#
#The following people and organisations have contributed to the development of the database in various ways. It is their work that makes it all possible.
#
# Korin Richmond
# Rob Clark
# Oliver Watts
# Chris Pidcock
# Graham Leary
# Blaise Potard
# Adevarul Online
# DEX Online - Catalin Francu
# Paul Borza
# Ovidiu Sabou
# Doina Tatar
# Mircea Giurgiu
# European Social Fund POSDRU/6/1.5/S/5
#
#and others too.
#
#
# This script for Hugging Face's datasets library was written by Théo Gigant


import csv
import json
import os

import datasets

_CITATION = """\
@article{Stan2011442,
  author = {Adriana Stan and Junichi Yamagishi and Simon King and
                   Matthew Aylett},
  title = {The {R}omanian speech synthesis ({RSS}) corpus:
                   Building a high quality {HMM}-based speech synthesis
                   system using a high sampling rate},
  journal = {Speech Communication},
  volume = {53},
  number = {3},
  pages = {442--450},
  note = {},
  abstract = {This paper first introduces a newly-recorded high
                   quality Romanian speech corpus designed for speech
                   synthesis, called ''RSS'', along with Romanian
                   front-end text processing modules and HMM-based
                   synthetic voices built from the corpus. All of these
                   are now freely available for academic use in order to
                   promote Romanian speech technology research. The RSS
                   corpus comprises 3500 training sentences and 500 test
                   sentences uttered by a female speaker and was recorded
                   using multiple microphones at 96 kHz sampling
                   frequency in a hemianechoic chamber. The details of the
                   new Romanian text processor we have developed are also
                   given. Using the database, we then revisit some basic
                   configuration choices of speech synthesis, such as
                   waveform sampling frequency and auditory frequency
                   warping scale, with the aim of improving speaker
                   similarity, which is an acknowledged weakness of
                   current HMM-based speech synthesisers. As we
                   demonstrate using perceptual tests, these configuration
                   choices can make substantial differences to the quality
                   of the synthetic speech. Contrary to common practice in
                   automatic speech recognition, higher waveform sampling
                   frequencies can offer enhanced feature extraction and
                   improved speaker similarity for HMM-based speech
                   synthesis.},
  doi = {10.1016/j.specom.2010.12.002},
  issn = {0167-6393},
  keywords = {Speech synthesis, HTS, Romanian, HMMs, Sampling
                   frequency, Auditory scale},
  url = {http://www.sciencedirect.com/science/article/pii/S0167639310002074},
  year = 2011
}
"""

_DESCRIPTION = """\
The Romanian speech synthesis (RSS) corpus was recorded in a hemianechoic chamber (anechoic walls and ceiling; floor partially anechoic) at the University of Edinburgh. We used three high quality studio microphones: a Neumann u89i (large diaphragm condenser), a Sennheiser MKH 800 (small diaphragm condenser with very wide bandwidth) and a DPA 4035 (headset-mounted condenser). Although the current release includes only speech data recorded via Sennheiser MKH 800, we may release speech data recorded via other microphones in the future. All recordings were made at 96 kHz sampling frequency and 24 bits per sample, then downsampled to 48 kHz sampling frequency. For recording, downsampling and bit rate conversion, we used ProTools HD hardware and software. We conducted 8 sessions over the course of a month, recording about 500 sentences in each session. At the start of each session, the speaker listened to a previously recorded sample, in order to attain a similar voice quality and intonation.
"""

_HOMEPAGE = "http://romaniantts.com/rssdb/"

_LICENSE = "CCPL"

_URLS = {
    "ro": "RomanianDB_v.0.8.1.tgz",
}

class RomanianSpeechSynthesis(datasets.GeneratorBasedBuilder):

    VERSION = datasets.Version("0.8.1")

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(name="ro", version=VERSION, description=""),
    ]

    DEFAULT_CONFIG_NAME = "ro"

    def _info(self):
        features = datasets.Features(
            {
                "sentence": datasets.Value("string"),
                "audio": datasets.features.Audio(sampling_rate=48_000),
            }
        )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        urls = _URLS[self.config.name]
        data_dir = dl_manager.download_and_extract(urls)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "datapath": data_dir,
                    "split": "training",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "datapath": data_dir,
                    "split": "testing"
                },
            ),
        ]

    def _generate_examples(self, datapath, split):
        key = 0
        audio_folder = "wav"
        for folder in (os.listdir(os.path.join(datapath, split, audio_folder))):
            with open(os.path.join(datapath, split, "text", folder+".txt")) as text_file:
               for line in text_file.readlines():
                    i = line[:3]
                    filename = f"adr_{folder}_{i}.wav"
                    local_path = os.path.join(split, audio_folder, folder, filename)
                    yield key, {
                      "sentence": line[4:-1],
                      "audio" : os.path.join(datapath, local_path)
                    }
                    key += 1