text_to_speech.md

Text To Speech

VLSP 2018 Shared Task: Aspect Text-To-Speech Evaluation

• 📜 VLSP 2018 Shared Task: Aspect Text-To-Speech Evaluation paper

In order to evaluate the quality of TTS systems, the test set contains 30 numbered sentences in the news domain. These sentences have different length, and contain some information on date, personal name, foreign location name, and some Vietnamese popular abbreviations, etc. The test set are collected from titles of e-newspapers such as dantri.com.vn, vnexpress.net, vietnamnet.vn, etc. on March 7 th , 2018. The TTS test set is released on March 8 th , 2018 at 8:30am. Each registered team can choose to generate speech in one or more dialects of the three regions: Northern, Southern and Central.

Leaderboard

Model	Score			Paper/Source	Code
	MOS	Naturalness	Intelligibility
Viettel	4.66	91.58	94.18	Nguyen et al. VLSP'18
MICA	3.90	76.06	82.10	Nguyen et al. VLSP'18
VAIS	3.46	65.96	72.51	Do et al. VLSP'18

Miscellaneous

📁 Open sources

• https://github.com/TensorSpeech/TensorFlowTTS
• https://github.com/NTT123/vietTTS
• https://github.com/NTT123/Vietnamese-Text-To-Speech-Dataset
• Smart Meeting Diarization - Smart Meeting Diarization by AILab (2017) demo
• VOS - Voice of Southern Vietnam by AILab (2017) demo
• vaisvn/hts_for_vietnamese - HTS for Vietnamese by VAIS (2016) perl,bash data

💫 Services

• vbee.vn (2018-)
• vtcc.ai TTS (2018-)
• OpenFPT: Speech Synthesis (2017-)
• ResponsiveVoice (2017)

📁 Dataset

• VIVOS - a free Vietnamese speech corpus consisting of 15 hours of recording speech by AILab (2017) data
• InfoRe1 - a public access speech corpus consisting of 25 hours of recording speech by InfoRe (2019) data