styletts2-kokoro-82m

A StyleTTS2 model by hexgrad, from hexgrad/Kokoro-82M, packaged for phoonnx, our ONNX text-to-speech library.

Model details

Architecture	StyleTTS2
Upstream	hexgrad/Kokoro-82M
ONNX conversion	onnx-community/Kokoro-82M-v1.0-ONNX
Author	hexgrad
Languages	en-GB, en-US, es, fr, hi, it, ja-JP, pt-BR, zh
Text input	espeak phonemes, IPA alphabet; misaki_en phonemes, IPA alphabet; misaki_ja phonemes, IPA alphabet; misaki_zh phonemes, IPA alphabet

How to use

pip install --pre phoonnx "scriptconv[misaki]"

The phonemes come from the espeak-ng program, so install it too (for example apt install espeak-ng). Without it, pip install --pre "phoonnx[espeak]" gives phoonnx the espyak port of espeak-ng. Some espyak phonemes differ from those of espeak-ng, so the voice can say some words wrong with espyak.

import wave
from phoonnx.model_manager import TTSModelManager

manager = TTSModelManager()
manager.load()
manager.merge_default_voices()

voice = manager.voices["kokoro/af"].load()

with wave.open("output.wav", "wb") as wav_file:
    text = "..."  # a sentence in en-US
    voice.synthesize_wav(text, wav_file)

In OVOS, phoonnx installs the ovos-tts-plugin-phoonnx plugin. Select the voice in mycroft.conf:

{"tts": {"module": "ovos-tts-plugin-phoonnx", "ovos-tts-plugin-phoonnx": {"voice": "kokoro/af"}}}

Text front end

phoonnx hands the text to scriptconv, our text conversion library, which normalises it and turns it into the model's input:

• espeak: EspeakPhonemizer from scriptconv.phonemizers.mul, installed by the scriptconv[espeak] extra.
• misaki_en: MisakiEnPhonemizer from scriptconv.phonemizers.mul, installed by the scriptconv[misaki] extra.
• misaki_ja: MisakiJaPhonemizer from scriptconv.phonemizers.mul, installed by the scriptconv[misaki] extra.
• misaki_zh: MisakiZhPhonemizer from scriptconv.phonemizers.mul, installed by the scriptconv[misaki] extra.

phoonnx voices

• kokoro/af
• kokoro/af_alloy
• kokoro/af_aoede
• kokoro/af_bella
• kokoro/af_heart
• kokoro/af_jessica
• kokoro/af_kore
• kokoro/af_nicole
• kokoro/af_nova
• kokoro/af_river
• kokoro/af_sarah
• kokoro/af_sky
• kokoro/am_adam
• kokoro/am_echo
• kokoro/am_eric
• kokoro/am_fenrir
• kokoro/am_liam
• kokoro/am_michael
• kokoro/am_onyx
• kokoro/am_puck
• kokoro/am_santa
• kokoro/bf_alice
• kokoro/bf_emma
• kokoro/bf_isabella
• kokoro/bf_lily
• kokoro/bm_daniel
• kokoro/bm_fable
• kokoro/bm_george
• kokoro/bm_lewis
• kokoro/ef_dora
• kokoro/em_alex
• kokoro/em_santa
• kokoro/ff_siwis
• kokoro/hf_alpha
• kokoro/hf_beta
• kokoro/hm_omega
• kokoro/hm_psi
• kokoro/if_sara
• kokoro/im_nicola
• kokoro/jf_alpha
• kokoro/jf_gongitsune
• kokoro/jf_nezumi
• kokoro/jf_tebukuro
• kokoro/jm_kumo
• kokoro/pf_dora
• kokoro/pm_alex
• kokoro/pm_santa
• kokoro/zf_xiaobei
• kokoro/zf_xiaoni
• kokoro/zf_xiaoxiao
• kokoro/zf_xiaoyi
• kokoro/zm_yunjian
• kokoro/zm_yunxi
• kokoro/zm_yunxia
• kokoro/zm_yunyang

Licence

The upstream declares the licence apache-2.0 in the model card of hexgrad/Kokoro-82M.

Related projects

• phoonnx: runs this model, in Python and in OVOS.
• scriptconv: text normalisation, script conversion and phonemizers for phoonnx.
• The StyleTTS2 collection: every StyleTTS2 model packaged for phoonnx.