Looking for audio data set for English words

Question

I am trying to obtain an audio data set for a list of English words. The list doesn't have to be extensive (for example, the data set can only have four or five words), but each word should have more than 10 WAV files (large repetition with different waveforms).

Answer 1

Lingua Libre has a lot of pronunciation recordings, including in English, by many speakers:

• https://lingualibre.org
• https://commons.wikimedia.org/wiki/Category:Lingua_Libre_pronunciation-eng
• https://lingualibre.org/datasets/

Unfortunately I have not found an easy way to find out what words have the most pronunciations. You might have to download the dataset and write a small script to find them.

Content is available under Creative Commons 4.0 Attribution-ShareAlike.

---

[Easy to use but not open data] Forvo is a crowdsourced effort to create sound files for every word of every language.

The great thing is that a word in a language can have more than 1 sound file.
In fact, you will often find sound files created by males and females from various regions with different accents and voices.

Pronunciations are ranked by clarity, so you can focus only on clear sound files, or on the contrary include difficult-to-understand sound files, depending on your goal.

• bread: 10 pronunciations from USA, UK, Malaysia
• anything: 30 pronunciations

This page contains the words that have most pronunciations:
http://www.forvo.com/languages-pronunciations/en/by-popularity/

Unfortunately the data can be reused but not for commercial purposes.

Answer 2

1. Pete Warden from Google organised efforts around a simple set of command words that might fit your needs (i.e., you wanted lots of variety even if the range of words was somewhat limited). They cover the digits plus: On, Off, Stop, Go, Up, Down, Left, Right, Yes, and No.

   Background details: https://arxiv.org/pdf/1804.03209.pdf / https://github.com/petewarden/open-speech-recording

   Dataset: http://download.tensorflow.org/data/speech_commands_v0.02.tar.gz

2. If you want a wider set, it would need a little more work to organise in the manner you were after, but a much larger set (En: ~22 GB) has been produced by Mozilla with their Common Voice project which gathers a large range of phrases spoken by individuals with various regional accents.

   Approaches may vary, but you could grep the labels for the word/phrases of interest and then use a force aligner tool like Gentle to get the particular words desired out of the corresponding .wav files.

   Background details: https://voice.mozilla.org/en

   Dataset: https://voice.mozilla.org/en/datasets