The train natchatts on my own data ,synthesis speech is silence.

[zhaojingxin123]

Use the speech of GenShin inside the KeQing (a game character) train the matcha_tts model , and use the trained model synthesis speech which is silence.
but the synthesis speech have mel spectrogram。
the synthesised mel spectrogram picture

Has this ever happened to you?What do you think is the cause?
thank you .

[shivammehta25]

Could you try replacing the vocoder with BigVGAN or even Griffin-lim for testing? I haven't faced this issue before; usually, the hifigan works just fine for the audio I have tested it with.

[shivammehta25]

Also, could you just verify that the sample rate of the input audio is 22050?

Matcha-TTS/configs/data/ljspeech.yaml

Line 13 in 108906c

sample_rate: 22050