schaltung / fae-cv

This repo contains the reicpe to assemble a corpus for Foreign Accented English using the crowdsourced corpus Common Voice which contains (optional) accent labels.

FAE Common Voice 2022

Foreign-Accented English from the crowdsourced Common Voice corpus.

Design Criteria:

• Use validated.tsv recordings only from cv-corpus-10.0-2022-07-04.
• ~20sec per speaker.
• ~100-500 speakers per accent-class.
• Favor speaker diversity and gender balance.

---

Requirements.

• Install docker-compose version > 1.29 or Compose V2 preferrably.
• Create a .env file and define these variable:
  • PATH_CORPORA=/local/directory/where/cv/was/downloaded/
  • PORT_JUPY=8888 # local port of choice

For example .env:

PATH_CORPORA=/data/corpora/CommonVoice
PORT_JUPY=0  # 0 picks random port

---

Build docker image.

Note: If using Compose V2, simply replace docker-compose with docker compose.

docker-compose build

Run docker container.

docker-compose -p fae$USER up -d --force-recreate

Running Jupyter notebook

Find the assigned local port by inspecting

docker-compose -p fae$USER ps
       Name                    Command                 State                         Ports                   
-------------------------------------------------------------------------------------------------------------
faejsmith_FAECV_1   tini -g -- start-notebook.sh   Up (healthy)   0.0.0.0:49153->8888/tcp,:::49153->8888/tcp

The container is listening to that port: http://server:49153/

Find the corresponding access token in the container logs:

docker-compose -p fae$USER logs

Sample visualization of the Accents-Embedding space.

Follow the FAE/CV22 recipe to duplicate the visualization below.

References

[1] Ardila R, Branson M, Davis K, Henretty M, Kohler M, Meyer J, Morais R, Saunders L, Tyers FM, Weber G. (2019). Common voice: A massively-multilingual speech corpus. arXiv preprint arXiv:1912.06670. 2019 Dec 13.