TextToSpeech

Text To Speech (ailia SDK for Unity)

This folder provides neural TTS pipelines. The scene is TextToSpeech/TextToSpeechSample.unity, controller TextToSpeech/AiliaVoiceSample.cs.

Supported Models

• Tacotron2 + WaveGlow (English)

  • Purpose: Sequence‑to‑mel + vocoder synthesis.
  • Source: ./AiliaVoiceSample.cs
  • Script behavior: Opens encoder/decoder/postnet/waveglow ONNX and synthesizes speech from text.

• GPT‑SoVITS (Japanese/English/Chinese)

  • Purpose: High‑fidelity TTS with SoVITS and GPT modules.
  • Source: ./AiliaVoiceSample.cs
  • Script behavior: Opens T2S encoder/FS decoder/decoder + VITS + cnhubert encoder; uses G2P for JA/EN/ZH.

• GPT‑SoVITS v2 (Japanese/English/Chinese)

  • Purpose: Improved fidelity TTS with SoVITS v2 and GPT modules.
  • Source: ./AiliaVoiceSample.cs
  • Script behavior: Opens T2S encoder/FS decoder/decoder + VITS + cnhubert encoder; uses G2P for JA/EN/ZH. Optional Chinese BERT support.

• GPT‑SoVITS v3 (Japanese/English/Chinese)

  • Purpose: High‑fidelity TTS with SoVITS v3, VQ, CFM, and BigVGAN modules.
  • Source: ./AiliaVoiceSample.cs
  • Script behavior: Opens T2S encoder/FS decoder/decoder + cnhubert + VQ + CFM + BigVGAN; uses G2P for JA/EN/ZH. Optional Chinese BERT support.

• GPT‑SoVITS v2 Pro (Japanese/English/Chinese)

  • Purpose: Enhanced TTS with SoVITS v2 Pro, VITS, and SV modules.
  • Source: ./AiliaVoiceSample.cs
  • Script behavior: Opens T2S encoder/FS decoder/decoder + cnhubert + VITS + SV; uses G2P for JA/EN/ZH. Optional Chinese BERT support.

• GPT‑SoVITS v2 Pro Distill (Japanese)

  • Purpose: Distilled (lightweight) version of v2 Pro for faster inference.
  • Source: ./AiliaVoiceSample.cs
  • Script behavior: Uses distilled T2S encoder/FS decoder/decoder with shared v2 Pro cnhubert/VITS/SV models; uses G2P for JA.

Model Files and Paths

• Models download into Application.temporaryCachePath under tacotron2, gpt-sovits, gpt-sovits-v2, gpt-sovits-v3, gpt-sovits-v2-pro, gpt-sovits-v2-pro-distill, g2p_en, g2p_cn, and g2pw.

Reusing in Your Own Code

• Initialize AiliaVoice pipeline:
  • For Tacotron2: OpenModel(encoder, decoder_iter, postnet, waveglow, null, AILIA_VOICE_MODEL_TYPE_TACOTRON2, ...).
  • For GPT‑SoVITS v1: OpenGPTSoVITSV1ModelFile(t2s_encoder, t2s_fsdec, t2s_sdec, vits, cnhubert).
  • For GPT‑SoVITS v2: OpenGPTSoVITSV2ModelFile(t2s_encoder, t2s_fsdec, t2s_sdec, vits, cnhubert, chinese_bert, vocab).
  • For GPT‑SoVITS v3: OpenGPTSoVITSV3ModelFile(t2s_encoder, t2s_fsdec, t2s_sdec, cnhubert, vq, cfm, bigvgan, chinese_bert, vocab).
  • For GPT‑SoVITS v2 Pro: OpenGPTSoVITSV2ProModelFile(t2s_encoder, t2s_fsdec, t2s_sdec, cnhubert, vits, sv, chinese_bert, vocab).
• Optional G2P:
  • Japanese: G2P(text, AILIA_VOICE_G2P_TYPE_GPT_SOVITS_JA); English: ..._EN; Chinese: ..._ZH.

Source Links

• Controller: ./AiliaVoiceSample.cs