search

keithito / tacotron

A TensorFlow implementation of Google's Tacotron speech synthesis with pre-trained model (unofficial)
MIT License
2.96k stars 956 forks source link

Repetition at end of synthesised wavs #326

Open fathimad11 opened 4 years ago

fathimad11 commented 4 years ago

I am using 4 min audio data which is converted to the same format askeithio did. By using transfer learning I reached about 7 lakh steps. The synthesised wavs are quite clear but contionus repetition at end. Help me please

ghost commented 4 years ago

Probably the data yo are using is quite less. have you trained it using LJspeech english dataset

fathimad11 commented 4 years ago

I used pretrained model provide by keithio used that as checkpoint and change my dataset to 4min male speaking in english.

On Thu, Mar 19, 2020 at 6:05 AM saharsyed notifications@github.com wrote:

Probably the data yo are using is quite less. have you trained it using LJspeech english dataset

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/keithito/tacotron/issues/326#issuecomment-601093217, or unsubscribe https://github.com/notifications/unsubscribe-auth/AONUIF7OQ2NJKOEKDGNGH3TRIHU6ZANCNFSM4LPE4R6Q .

ghost commented 4 years ago

i assume 4 minutes is way too less