Vocoder with HiFIGAN trained on custom German dataset

This repository provides all the necessary tools for using a HiFIGAN vocoder trained on a generated German dataset using mp3_to_training_data.
The pre-trained model (8 epochs so far) takes in input a spectrogram and produces a waveform in output. Typically, a vocoder is used after a TTS model that converts an input text into a spectrogram.

How to use

Install speechbrain.
pip install speechbrain<br />

Use a TTS model (e.g. tts-tacotron-german), generate a spectrogram and convert it to audio.
import torchaudio<br /> from speechbrain.pretrained import Tacotron2<br /> from speechbrain.pretrained import HIFIGAN<br /> tacotron2 = Tacotron2.from_hparams(source="padmalcom/tts-tacotron2-german", savedir="tmpdir_tts")<br /> hifi_gan = HIFIGAN.from_hparams(source="padmalcom/tts-hifigan-german", savedir="tmpdir_vocoder")<br /> mel_output, mel_length, alignment = tacotron2.encode_text("Mary had a little lamb")<br /> waveforms = hifi_gan.decode_batch(mel_output)<br /> torchaudio.save('example_TTS.wav',waveforms.squeeze(1), 22050)<br />

Inference on GPU

To perform inference on the GPU, add run_opts={"device":"cuda"} when calling the from_hparams method.