Results

Leaderboard

The leaderboard above presents the results obtained by the participants to the 2019 edition. The columns are sortable by clicking on the picture of each column header. A detailed view of the results is available by clicking on the picture of each row, it includes audio samples of speech synthesis.

The score columns are interpreted as follow (see Evaluation metrics for more details):

• MOS:

  • mean opinion score on speech synthesis

  • scale is \([1, 5]\), bigger is better

• CER:

  • character error rate after human transcription of speech synthesis

  • scale is \([0, 1]\), lower is better

• Similarity:

  • similarity to the target voice of speech synthesis

  • scale is \([1, 5]\), bigger is better

• ABX:

  • ABX error rate on embeddings

  • scale is \([0, 100]\), lower is better

• Bitrate:

  • bitrate of the embeddings

  • scale is \(]0, +\infty[\), lower is better

	#	Authors	Surprise language					Training language (English)
			MOS	CER	Similarity	ABX	Bitrate	MOS	CER	Similarity	ABX	Bitrate
	#	Authors	MOS	CER	Similarity	ABX	Bitrate	MOS	CER	Similarity	ABX	Bitrate
			Surprise language					Training language (English)

Graphs

MOS/Bitrate CER/Bitrate Similarity/Bitrate ABX/Bitrate CER/MOS MOS/ABX CER/ABX