Skip to content
ModelListedExternal
Browse files

Irodori Tts GGUF

GGUF / ggml conversion of Aratako/Irodori-TTS-500M-v3 for use with CrispStrobe/CrispASR.

Sourcehuggingface.co/cstr/irodori-tts-GGUF/tree/0c961a42ed181c27a129803e6d3d572e3f4eb892
KindOtherBase modelAratako/Irodori-TTS-500M-v3VersionRevision 0c961a4Bgrade

Model source

Kind
Other
File size
7.88 KB–979 MB
Base model
Aratako/Irodori-TTS-500M-v3
Version
vRevision 0c961a4
Updated
Oct 31, 2026
Downloads
1.3K
Source
Hugging Face

Source description

Irodori-TTS-500M-v3 — GGUF (ggml-quantised)

GGUF / ggml conversion of Aratako/Irodori-TTS-500M-v3 for use with CrispStrobe/CrispASR.

Irodori-TTS is a 500M-param Rectified Flow Diffusion Transformer (RF-DiT) text-to-speech model with zero-shot voice cloning via DAC-VAE latent conditioning. Japanese-focused with a 102K-vocab sarashina2.2 tokenizer. 48 kHz mono output.

License: MIT (follows upstream Irodori-TTS license).

Files

FileQuantSizeNotes
irodori-tts-500m-v3-f16.ggufF16~1.9 GBReference quality
irodori-tts-500m-v3-q4_k.ggufQ4_K~852 MBRecommended — fits 8 GB RAM
irodori-tts-500m-v3-q8_0.ggufQ8_0~896 MBNear-lossless
irodori-tts-ref.ggufF32~4 KBReference activations for diff harness

Architecture

Text Input (Japanese / mixed)
  │
sarashina2.2 Tokenize (102K vocab, BPE)
  │
TextEncoder (14L, 1280d, 10 heads, RoPE + SwiGLU)
  │── Each position: self-attention + gated residual + SwiGLU FFN
  │
  ├── [Optional] ReferenceLatentEncoder (14L, 1280d)
  │    └── DAC-VAE latent from reference audio → speaker conditioning
  │
DiT Backbone (24L, 2048d, 16 heads)
  │── LowRankAdaLN (rank=256) timestep conditioning
  │── JointAttention: self-KV + text-context-KV + speaker-context-KV
  │── Half-RoPE (first half of head_dim rotated, rest passthrough)
  │── SwiGLU MLP (ratio 2.875)
  │
Euler RF ODE Solver (40 steps, CFG)
  │── noise → DAC-VAE latent sequence (32-dim continuous)
  │
Semantic-DACVAE Decoder (48 kHz reconstruction)
  │── 32-dim latent → Snake1d + ConvTranspose1d upsampling → PCM
  │
Output: float32 mono @ 48 kHz

Source model

Provenance and EU AI Act Art. 53 note

  • Upstream model: Aratako/Irodori-TTS-500M-v3 — published by Aratako.
  • Upstream licence: mit. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
  • What was done here: format conversion and/or quantisation only (GGUF/GGML). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
  • Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository.
  • Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.

Sources

1 source
Hugging FacePrimaryVersion 0c961a42ed181c27a129803e6d3d572e3f4eb892
huggingface.co/cstr/irodori-tts-GGUF/tree/0c961a42ed181c27a129803e6d3d572e3f4eb892

Verified Oct 31

Download files

Repository files

File inventory only; completeness is unverified.

Model artifacts

8 artifacts

dacvae-ja-32dim-f16.gguf

gguf · 187 MB · SHA-256 837df284f23d…4135 · Hugging Face

Download file

irodori-tts-500m-v3-f16.gguf

gguf · 979 MB · SHA-256 2f38d35b934f…5d0e · Hugging Face

Download file

irodori-tts-500m-v3-q4_k.gguf

gguf · Q4_K (unverified) · 389 MB · SHA-256 188126c5813b…bc03 · Hugging Face

Download file

irodori-tts-500m-v3-q8_0.gguf

gguf · Q8_0 (unverified) · 594 MB · SHA-256 21a96241aaa7…f37d · Hugging Face

Download file

irodori-tts-ref-stages.gguf

gguf · 168 KB · SHA-256 430ff9d98cad…dc52 · Hugging Face

Download file

irodori-tts-ref.gguf

gguf · 50.3 KB · Hugging Face

python_cfg_latent.bin

bin · 7.88 KB · SHA-256 4458f697eaf9…3284 · Hugging Face

Download file

python_nocfg_latent.bin

bin · 7.88 KB · SHA-256 8fce286f2472…e046 · Hugging Face

Download file

Source excerpts

3 excerpts
ReadmeAug 15huggingface.co/cstr/irodori-tts-GGUF

--- license: mit language: - ja base_model: - Aratako/Irodori-TTS-500M-v3 pipeline_tag: text-to-speech tags: - tts - text-to-speech - irodori-tts - flow-matching - diffusion-transformer - dacvae - gguf - crispasr - japanese library_name: ggml --- # Irodori-TTS-500M-v3 — GGUF (ggml-quantised) GGUF / ggml conversion of [`Aratako/Irodori-TTS-500M-v3`](https://huggingface.co/Aratako/Irodori-TTS-500M-v3) for use with **[CrispStrobe/CrispASR](https://github.com/CrispStrobe/CrispASR)**. Irodori-TTS is a 500M-param **Rectified Flow Diffusion Transformer (RF-DiT)** text-to-speech model with zero-shot voice cloning via DAC-VAE latent conditioning. Japanese-focused with a 102K-vocab sarashina2.2 tokenizer. 48 kHz mono output. **License:** MIT (follows upstream Irodori-TTS license). ## Files | File | Quant | Size | Notes | |---|---|---:|---| | `irodori-tts-500m-v3-f16.gguf` | F16 | ~1.9 GB | Reference quality | | `irodori-tts-500m-v3-q4_k.gguf` | Q4_K | ~852 MB | **Recommended** — fits 8 GB RAM | | `irodori-tts-500m-v3-q8_0.gguf` | Q8_0 | ~896 MB | Near-lossless | | `irodori-tts-ref.gguf` | F32 | ~4 KB | Reference activations for diff harness | ## Architecture ``` Text Input (Japanese / mixed) │ sarashina2.2 Tokenize (102K vocab, BPE) │ TextEncoder (14L, 1280d, 10 heads, RoPE + SwiGLU) │── Each position: self-attention + gated residual + SwiGLU FFN │ ├── [Optional] ReferenceLatentEncoder (14L, 1280d) │ └── DAC-VAE latent from reference audio → speaker conditioning │ DiT Backbone (24L, 2048d, 16 heads) │── LowRankAdaLN (rank=256) timestep conditioning │── JointAttention: self-KV + text-context-KV + speaker-context-KV │── Half-RoPE (first half of head_dim rotated, rest passthrough) │── SwiGLU MLP (ratio 2.875) │ Euler RF ODE Solver (40 steps, CFG) │── noise → DAC-VAE latent sequence (32-dim continuous) │ Semantic-DACVAE Decoder (48 kHz reconstruction) │── 32-dim latent → Snake1d + ConvTranspose1d upsampling → PCM │ Output: float32 mono @ 48 kHz ``` ## Source model - **Upstream:** [Aratako/Irodori-TTS-500M-v3](https://huggingface.co/Aratako/Irodori-TTS-500M-v3) (safetensors, ~1.9 GB) - **Codec:** [Aratako/Semantic-DACVAE-Japanese-32dim](https://huggingface.co/Aratako/Semantic-DACVAE-Japanese-32dim) - **Tokenizer:** [sbintuitions/sarashina2.2-0.5b](https://huggingface.co/sbintuitions/sarashina2.2-0.5b) - **Code:** [Aratako/Irodori-TTS](https://g...

NoteOct 31huggingface.co/cstr/irodori-tts-GGUF

Source context: 1258 downloads · 1 likes · Pipeline text-to-speech · Library ggml · Repo cstr/irodori-tts-GGUF

NoteAug 15huggingface.co/cstr/irodori-tts-GGUF

Source context: 947 downloads · 1 likes · Pipeline text-to-speech · Library ggml · Repo cstr/irodori-tts-GGUF