import tempfile
from typing import Optional
from TTS.config import load_config
import gradio as gr
import numpy as np
from TTS.utils.manage import ModelManager
from TTS.utils.synthesizer import Synthesizer
MODELS = {}
SPEAKERS = {}
MAX_TXT_LEN = 100
manager = ModelManager()
MODEL_NAMES = manager.list_tts_models()
# filter out multi-speaker models and slow wavegrad vocoders
filters = ["vctk", "your_tts", "ek1"]
MODEL_NAMES = [model_name for model_name in MODEL_NAMES if not any(f in model_name for f in filters)]
EN = [el for el in MODEL_NAMES if "/en/" in el]
OTHER = [el for el in MODEL_NAMES if "/en/" not in el]
EN[0], EN[5] = EN[5], EN[0]
MODEL_NAMES = EN + OTHER
# reorder models
print(MODEL_NAMES)
def tts(text: str, model_name: str, speaker_idx: str=None):
if len(text) > MAX_TXT_LEN:
text = text[:MAX_TXT_LEN]
print(f"Input text was cutoff since it went over the {MAX_TXT_LEN} character limit.")
print(text, model_name)
# download model
model_path, config_path, model_item = manager.download_model(model_name)
vocoder_name: Optional[str] = model_item["default_vocoder"]
# download vocoder
vocoder_path = None
vocoder_config_path = None
if vocoder_name is not None:
vocoder_path, vocoder_config_path, _ = manager.download_model(vocoder_name)
# init synthesizer
synthesizer = Synthesizer(
model_path, config_path, None, None, vocoder_path, vocoder_config_path,
)
# synthesize
if synthesizer is None:
raise NameError("model not found")
wavs = synthesizer.tts(text, speaker_idx)
# return output
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
synthesizer.save_wav(wavs, fp)
return fp.name
article= """
Visit us on Coqui.ai and drop a π to πCoquiTTS.
Run CoquiTTS locally for the best result. Check out our πdocumentation.
```bash
$ pip install TTS
...
$ tts --list_models
...
$ tts --text "Text for TTS" --model_name "///" --out_path folder/to/save/output.wav
```
π Model contributors
- @nmstoker
- @kaiidams
- @WeberJulian,
- @Edresson
- @thorstenMueller
- @r-dh
- @kirianguiller
- @robinhad
- @fkarabiber
- @nicolalandro
- @a-froghyar
π Drop a β¨PRβ¨ on πΈTTS to share a new model and have it included here.
"""
iface = gr.Interface(
fn=tts,
inputs=[
gr.inputs.Textbox(
label="Input Text",
default="This sentence has been generated by a speech synthesis system.",
),
gr.inputs.Radio(
label="Pick a TTS Model - (language/dataset/model_name)",
choices=MODEL_NAMES,
),
# gr.inputs.Dropdown(label="Select a speaker", choices=SPEAKERS, default=None)
# gr.inputs.Audio(source="microphone", label="Record your voice.", type="numpy", label=None, optional=False)
],
outputs=gr.outputs.Audio(label="Output"),
title="πΈπ¬ CoquiTTS Demo",
theme="grass",
description="πΈπ¬ Coqui TTS - a deep learning toolkit for Text-to-Speech, battle-tested in research and production.",
article=article,
allow_flagging=False,
flagging_options=['error', 'bad-quality', 'wrong-pronounciation'],
layout="vertical",
live=False
)
iface.launch(share=False)