myshell-ai
Fix Loudness Issue
Tested with Chinese and found no clipping.
| PR | Waveform |
|---|---|
| Before | |
| After |
Text used:
我要給阿Ｑ做正傳，已經不止一兩年了。但一面要做，一面又往回想，這足見我不是一個「立言」的人，因為從來不朽之筆，須傳不朽之人，於是人以文傳，文以人傳——究竟誰靠誰傳，漸漸的不甚瞭然起來，而終於歸接到傳阿Ｑ，彷彿思想裡有鬼似的。
Edit: Based on this PR https://github.com/myshell-ai/MeloTTS/pull/221, it normalizes the audio stream without affecting audio quality. However, the implementation is now different.
Hello, @alastorid
I'm source melo TTS recently, and try implement this model to edge devcie. Follow your PR, I meet below problem. Please have a time to help, thank you! In melo/app.py
- audio_list.append(utils.fix_loudness(audio,self.hps.data.sampling_rate))
cause almost audio content lost, just save the last around 3 seconds audio. base on aarch64/conda/python3.9/ubuntu22.04. All dependencies installed as the requirements.
Hello @HarryBXie , Thanks for reporting the issue. I’ve reworked the solution, and it now works without the need for additional dependencies, unlike before. Could you please pull the latest version of this PR and try again?
@alastorid, Thank you for your quick response, new PR is works well. I have tried two kinds of development platform, the PR both are effective.