ENST Camille Voice Data

A French speech corpus designed for synthesis, recorded in 2013 at Télécom ParisTech (ENST) by Camille Dianoux, a female native speaker of French.

Data format

Audio

The audio data is provided in the losslessly compressed FLAC format, which can be played by a myriad of software, including Praat. The speaker was recorded at a 44.1 kHz sampling rate, 16 bits per sample, in mono. No filters of any sort have been applied to this raw data.

Phonetic segmentation

Annotations are provided as a single YAML file. It contains a list of utterances, each of which consists of

• a prompt code (file basename),
• the utterance text,
• the recording date,
• utterance start and end times (in seconds) in the FLAC file,
• the phonetic segments (obtained using the eHMM tool from FestVox 2.1), each of which has
  • a label (based on SAMPA, _ denotes silence), and
  • its duration (in seconds)

For example,

- prompt: text_0366
  text: les alpinistes installent un bivouac au pied de la montagne .
  date: 2013-06-17T15:34:58Z
  start: 3185.5455782309
  end: 3191.814965986
  segments:
  - { lab: _, dur: 0.614969 }
  - { lab: l, dur: 0.08 }
  - { lab: e, dur: 0.135 }
  - { lab: a, dur: 0.09 }
  - { lab: l, dur: 0.045 }
  - { lab: p, dur: 0.085 }
  - { lab: i, dur: 0.075 }
  - { lab: n, dur: 0.06 }
  - { lab: i, dur: 0.105 }
  - { lab: s, dur: 0.095 }
  - { lab: t, dur: 0.065 }
  - { lab: '@', dur: 0.115 }
  # etc.

Extracting the corpus

Prerequisites

Java 8 (or later) and SoX must be installed.

Assembling the data

The data processing is delegated to Gradle and the FLAML plugin.

Run ./gradlew unpackData extractTextFiles extractLabFiles extractWavFiles to download and extract all data. See the FLAML plugin documentation for details.

Run ./gradlew assemble to prepare the data for distribution.

License

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. See the LICENSE.md file for details.