Common Voice

from Wikipedia, the free encyclopedia
Globe icon of the infobox
Common Voice
Website logo
"Teaching machines how real people speak"
Collection of spoken sentences for speech recognition - database
languages multilingual ( list of languages )
operator Mozilla Foundation
user over 20,000 (as of November 2017)
Registration Optional
On-line June 19, 2017 (currently active)
https://voice.mozilla.org/

Common Voice is a crowdsourcing project launched by Mozilla to create a free database for speech recognition software . The project is supported by volunteers who speak sample sentences with a microphone and review recordings made by other users. The transcribed sentences are collected in a voice database available under the public domain license CC0 . This ensures that developers can use the database for speech-to-text applications without restrictions or costs. Common Voice appeared as an answer to the voice assistants of large companies such as Amazon Echo , Siri or Google Assistant . Since June 6, 2018, German sentences can also be spoken and checked.

Naming

The name "Common Voice" comes from English and translates as "common voice". Mozilla chose the name based on Creative Commons , which is due to the common fundamental idea of ​​the common good.

Use in DeepSpeech

Parallel to Common Voice developed Mozilla, the speech recognition engine Deep Speech , a TensorFlow implementation services of the Deep Speech architecture of Baidu . Among other things, this uses the data from Common Voice and, according to its own statements, achieves a word error rate ( WER ) of 5.97 percent when it comes to recognizing American English when testing the LibriSpeech Clean Test Corpus. The human error rate is 5.83 percent. The source code of the fully trained neural network is available on GitHub . DeepSpeech is used by the free language assistant Mycroft , among others .

statistics

Common Voice's English database is now the largest freely accessible voice database ahead of LibriSpeech . When the last English dataset was published on December 10, 2019, over 50,000 users worldwide submitted 1,118 hours of validated sentences.

So far, more than 8,460 speakers have participated in the German database. The total of the German voice recordings confirmed by the volunteers amounts to over 483 hours. (As of December 10, 2019)

Related projects

  • LibriSpeech: Language corpus of around 1000 hours of English-language audio books from LibriVox (CC BY 4.0).
  • TED-LIUM: Language corpus of around 118 hours from English-language TED talks (CC BY-NC-ND 3.0).
  • Tatoeba : Collection of (spoken) example sentences in different languages ​​with translations (texts CC BY 2.0, audio mostly CC BY-NC-ND 3.0).
  • VoxForge: language corpus in different languages ​​to support free speech recognition engines like Julius (GPL 3).

Web links

Commons : Common Voice  - collection of pictures, videos and audio files

Individual evidence

  1. ^ Daniel Kessler: Raising Our Common Voice For The Web. In: The Mozilla Blog. June 19, 2017. Retrieved January 26, 2018 (American English).
  2. ↑ Frequently Asked Questions. In: Common Voice. Retrieved January 26, 2018 (American English).
  3. ↑ Wolfgang Reszel: Mozilla collects voice recordings for open speech recognition software. In: heise online. July 23, 2017, accessed January 26, 2018 .
  4. ↑ Sebastian Grüner: Common Voice: Mozilla's free language database becomes multilingual. In: Golem.de. June 7, 2018, accessed June 7, 2018 .
  5. ↑ Sebastian Grüner: Mozilla brings free speech recognition for everyone (page 2). In: Golem.de. November 30, 2017. Retrieved January 26, 2018 .
  6. ↑ Release Deep Speech 0.8.0. In: github.com. Mozilla, August 4, 2020, accessed August 4, 2020 .
  7. ↑ Sebastian Grüner: Mozilla brings free speech recognition for everyone (page 1). In: Golem.de. November 30, 2017. Retrieved January 26, 2018 .
  8. ↑ Stefan Bordel: Mozilla publishes huge speech dataset. In: com! The computer magazine. November 30, 2017. Retrieved January 26, 2018 .
  9. ↑ Steve Penrod: Why We're moving to DeepSpeech on March 31 . In: Mycroft (Ed.): Mycroft . January 11, 2018 ( mycroft.ai [accessed February 15, 2018]).
  10. ↑ Common Voice Records. December 10, 2019, accessed May 12, 2020 (American English).
  11. ↑ Herbert Braun: Mozilla Common Voice: Voice control for everyone and without resorting to the cloud. In: heise online. November 29, 2017. Retrieved January 26, 2018 .