Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -42,11 +42,12 @@ pipe_dict = {
|
|
42 |
"language": "english",
|
43 |
}
|
44 |
|
45 |
-
title =
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
|
|
50 |
|
51 |
max_speakers = 15
|
52 |
|
@@ -80,12 +81,22 @@ def generate_audio(text, language):
|
|
80 |
out.extend([gr.Audio(visible=False)]*(max_speakers-(len(out))))
|
81 |
return out
|
82 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
83 |
|
84 |
# Gradio blocks demo
|
85 |
-
with gr.Blocks() as demo_blocks:
|
86 |
gr.Markdown(title)
|
87 |
-
gr.
|
88 |
-
with gr.Row():
|
89 |
with gr.Column():
|
90 |
inp_text = gr.Textbox(label="Input Text", info="What would you like VITS to synthesise?")
|
91 |
btn = gr.Button("Generate Audio!")
|
@@ -101,6 +112,47 @@ with gr.Blocks() as demo_blocks:
|
|
101 |
for i in range(max_speakers):
|
102 |
out_audio = gr.Audio(type="numpy", autoplay=False, label=f"Generated Audio - speaker {i}", show_label=True, visible=False)
|
103 |
outputs.append(out_audio)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
|
105 |
btn.click(generate_audio, [inp_text, language], outputs)
|
106 |
|
|
|
42 |
"language": "english",
|
43 |
}
|
44 |
|
45 |
+
title = """# Explore English and Spanish Accents with VITS finetuning
|
46 |
+
## Or how the best wine comes in old bottles
|
47 |
+
[VITS](https://huggingface.co/docs/transformers/model_doc/vits) is a light weight, low-latency TTS model.
|
48 |
+
Coupled with the right datasets and the right training recipes, you can get an excellent finetuned version in 20 minutes with as little as 80 to 150 samples.
|
49 |
+
Stay tuned, the training recipe is coming soon!
|
50 |
+
""",
|
51 |
|
52 |
max_speakers = 15
|
53 |
|
|
|
81 |
out.extend([gr.Audio(visible=False)]*(max_speakers-(len(out))))
|
82 |
return out
|
83 |
|
84 |
+
css = """
|
85 |
+
#container{
|
86 |
+
margin: 0 auto;
|
87 |
+
max-width: 80rem;
|
88 |
+
}
|
89 |
+
#intro{
|
90 |
+
max-width: 100%;
|
91 |
+
text-align: center;
|
92 |
+
margin: 0 auto;
|
93 |
+
}
|
94 |
+
"""
|
95 |
|
96 |
# Gradio blocks demo
|
97 |
+
with gr.Blocks(css=css) as demo_blocks:
|
98 |
gr.Markdown(title)
|
99 |
+
with gr.Row(elem_id="container"):
|
|
|
100 |
with gr.Column():
|
101 |
inp_text = gr.Textbox(label="Input Text", info="What would you like VITS to synthesise?")
|
102 |
btn = gr.Button("Generate Audio!")
|
|
|
112 |
for i in range(max_speakers):
|
113 |
out_audio = gr.Audio(type="numpy", autoplay=False, label=f"Generated Audio - speaker {i}", show_label=True, visible=False)
|
114 |
outputs.append(out_audio)
|
115 |
+
|
116 |
+
|
117 |
+
gr.Markdown("""
|
118 |
+
## Datasets and models details
|
119 |
+
|
120 |
+
### English
|
121 |
+
|
122 |
+
* **Model**: [VITS-ljs](https://huggingface.co/kakao-enterprise/vits-ljs)
|
123 |
+
* **Dataset**: [British Isles Accent](https://huggingface.co/datasets/ylacombe/english_dialects). For each accent, we used 100 to 150 samples of a single speaker to finetune [VITS-ljs](https://huggingface.co/kakao-enterprise/vits-ljs).
|
124 |
+
|
125 |
+
### Spanish
|
126 |
+
|
127 |
+
* **Model**: [Spanish MMS TTS](https://huggingface.co/facebook/mms-tts-spa). This model is part of Facebook's [Massively Multilingual Speech](https://arxiv.org/abs/2305.13516) project, aiming to
|
128 |
+
provide speech technology across a diverse range of languages. You can find more details about the supported languages
|
129 |
+
and their ISO 639-3 codes in the [MMS Language Coverage Overview](https://dl.fbaipublicfiles.com/mms/misc/language_coverage_mms.html),
|
130 |
+
and see all MMS-TTS checkpoints on the Hugging Face Hub: [facebook/mms-tts](https://huggingface.co/models?sort=trending&search=facebook%2Fmms-tts).
|
131 |
+
* **Datasets**: For each accent, we used 100 to 150 samples of a single speaker to finetune the model.
|
132 |
+
- [Colombian Spanish TTS dataset](https://huggingface.co/datasets/ylacombe/google-colombian-spanish).
|
133 |
+
- [Argentinian Spanish TTS dataset](https://huggingface.co/datasets/ylacombe/google-argentinian-spanish).
|
134 |
+
- [Chilean Spanish TTS dataset](https://huggingface.co/datasets/ylacombe/google-chilean-spanish).
|
135 |
+
|
136 |
+
""")
|
137 |
+
|
138 |
+
with gr.Accordion("Run with transformers"):
|
139 |
+
gr.Markdown(
|
140 |
+
"""## Running VITS and MMS with transformers
|
141 |
+
```bash
|
142 |
+
pip install transformers
|
143 |
+
```
|
144 |
+
```py
|
145 |
+
from transformers import pipeline
|
146 |
+
import scipy
|
147 |
+
pipe = pipeline("text-to-speech", model="kakao-enterprise/vits-ljs", device=0)
|
148 |
+
|
149 |
+
results = pipe("A cinematic shot of a baby racoon wearing an intricate italian priest robe")
|
150 |
+
|
151 |
+
# write to a wav file
|
152 |
+
scipy.io.wavfile.write("audio_vits.wav", rate=results["sampling_rate"], data=results["audio"].squeeze())
|
153 |
+
```
|
154 |
+
"""
|
155 |
+
)
|
156 |
|
157 |
btn.click(generate_audio, [inp_text, language], outputs)
|
158 |
|