Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,40 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
---
|
4 |
+
|
5 |
+
This FastPitch[1] model was trained on the HUI-Audio-Corpus-German[2] clean dataset using the Nemo Toolkit[3].
|
6 |
+
We selected 5 speakers who have the 5-largest amount of data and balanced training data across speakers (around 20 hours per speaker).
|
7 |
+
|
8 |
+
|
9 |
+
|
10 |
+
This a retrained model of:
|
11 |
+
https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/tts_de_fastpitch_multispeaker_5
|
12 |
+
|
13 |
+
|
14 |
+
# How to Use:
|
15 |
+
Use with Nemo Toolkit
|
16 |
+
```python
|
17 |
+
# Load spectrogram generator
|
18 |
+
from nemo.collections.tts.models import FastPitchModel
|
19 |
+
spec_generator = FastPitchModel.restore_from("path/to/model.nemo")
|
20 |
+
|
21 |
+
# Load Vocoder
|
22 |
+
from nemo.collections.tts.models import HifiGanModel
|
23 |
+
model = HifiGanModel.from_pretrained(model_name="tts_de_hui_hifigan_ft_fastpitch_multispeaker_5")
|
24 |
+
|
25 |
+
# Generate audio
|
26 |
+
import soundfile as sf
|
27 |
+
parsed = spec_generator.parse("")
|
28 |
+
speaker_id = 0
|
29 |
+
spectrogram = spec_generator.generate_spectrogram(tokens=parsed, speaker=10)
|
30 |
+
audio = model.convert_spectrogram_to_audio(spec=spectrogram)
|
31 |
+
|
32 |
+
# Save the audio to disk in a file called speech.wav
|
33 |
+
sf.write("speech.wav", audio.to('cpu').numpy(), 44100)
|
34 |
+
```
|
35 |
+
|
36 |
+
|
37 |
+
|
38 |
+
[1] FastPitch: Parallel Text-to-speech with Pitch Prediction: https://arxiv.org/abs/2006.06873
|
39 |
+
[2] HUI-Audio-Corpus-German Dataset: https://opendata.iisys.de/datasets.html
|
40 |
+
[3] NVIDIA NeMo Toolkit: https://github.com/NVIDIA/NeMo
|