vctk-vits-onnx-cmudict

ESPnet VITS Text-to-Speech (TTS) Model for ONNX

espnet/kan-bayashi_vctk_vits exported to ONNX. This model is an ONNX export using the espnet_onnx library.

Usage with txtai

txtai has a built in Text to Speech (TTS) pipeline that makes using this model easy.

Note the following example requires txtai >= 7.5

import soundfile as sf

from txtai.pipeline import TextToSpeech

# Build pipeline
tts = TextToSpeech("NeuML/vctk-vits-onnx")

# Generate speech with speaker id
speech, rate = tts("Say something here", speaker=15)

# Write to file
sf.write("out.wav", speech, rate)

Usage with ONNX

This model can also be run directly with ONNX provided the input text is tokenized. Tokenization can be done with ttstokenizer.

Note that the txtai pipeline has additional functionality such as batching large inputs together that would need to be duplicated with this method.

import numpy as np
import onnxruntime
import soundfile as sf
import yaml

from ttstokenizer import TTSTokenizer

# This example assumes the files have been downloaded locally
with open("vctk-vits-onnx/config.yaml", "r", encoding="utf-8") as f:
    config = yaml.safe_load(f)

# Create model
model = onnxruntime.InferenceSession(
    "vctk-vits-onnx/model.onnx",
    providers=["CPUExecutionProvider"]
)

# Create tokenizer
tokenizer = TTSTokenizer(config["token"]["list"])

# Tokenize inputs
inputs = tokenizer("Say something here")

# Generate speech
outputs = model.run(None, {"text": inputs, "sids": np.array([15])})

# Write to file
sf.write("out.wav", outputs[0], 22050)

How to export

More information on how to export ESPnet models to ONNX can be found here.

Speaker reference

The CSTR VCTK Corpus includes speech data uttered by native speakers of English with various accents.

When using this model, set a speaker id from the reference table below. The ref column corresponds to the id in the VCTK dataset.