[Company] [Ivory Tower] PPG Ali paper tacotron part, draw the localization algorithm. Write the code PPG-TTS-Tacotron-ALi_pytorch, draw the table design hyperparameter search experiment . There is a preliminary result before January 31
[School] [Company] Mixed language stage2, bilingual corpus LID modified synthesis results sorting. PPG results are done after they come out, as the baseline ceiling
[Company] [Ivory Tower] summarized the results of the remaining three types of hyperparameter experiments , under the design λ, the influence of dim and wide on the VC capability, the design plan explores the plan that can be VC and rec_loss to the extreme (not easy to adjust, so the whole similar loss framework). After the PPG results come out, the LID BaseLine is also sorted out before doing
[School] Summarize the PPG Project method as a patent. The content of the PPT is organized into a blog , and write one according to the old patent, and expand it to CopyVC. Continue to do it after February 1st
[Company] BC2011 and other replies and Aishell1 are downloaded to the computer and uploaded to the intranet. Starting on January 26
[Company] Maybe-TTS is handed over according to the module, optimized and merged to calculate the py file of the pause text. A little bit every day
[Old School] Chinese speech processing paper reading, homework of the first graduate student, SLT paper reading, sorting of Xintao papers, the teacher recommends finding the nearest paper for reading and sorting, connecting with Mr. Ye’s melgan, and considering the Tacotron downmix text processing scheme based on the notes read the paper class. ML's look at the book. finishing various extraction methods and Lu Hui PPG and discussion. slowly
Task completed
[Company] [Ivory Tower] PPG Ali thesis version, PPG extracted documents, found that the server librispeech is still incomplete, I will not train the cpu version of ForcePPG, and I will consider it when training PPG in the future, and plan to communicate with Lu Hui. , Continue PPG-Tacotron code tomorrow
[Company] Data collection of BC2011 and Aishell1, and transfer to the company intranet, download in the bedroom
[Company] Maybe-TTS is handed over according to the module, optimized and merged to calculate the py file of the pause text, and will write next tomorrow
[Old School] Chinese speech processing paper reading, I got up late this morning, but I didn’t have time to read it. I’ll read it tomorrow