Irodori-TTS

Irodori-TTS is Japanese TTS under --family irodori_tts. It supports no-reference speech, optional reference-audio voice cloning, and instruction-based voice design.

The default downloadable package is the GGUF v4 Small Q8_0 checkpoint. v4 Small is the preferred path for new use because one checkpoint covers no-reference TTS, voice cloning, and voice design. The older v3 packages remain supported for existing users and local validation.

Variants

Variant	Package	Tasks	Notes
v4 Small	Irodori-TTS-v4-Small-GGUF	tts, clon, vdes	Bundles its v4 tokenizer and supports instruction/caption conditioning in the same checkpoint.
500M v3	Irodori-TTS-500M-v3-GGUF	tts, clon	Uses the shared llm-jp tokenizer layout in the original safetensors package.
600M v3 VoiceDesign	Irodori-TTS-600M-v3-VoiceDesign-GGUF	tts, clon, vdes	Adds instruction/caption conditioning for voice design.

v4 GGUF packages are published in both q8_0 and f16. v3 GGUF packages are also available in q8_0 and f16.

v4 reference-conditioning note: Fresh v4 voice-clone or reference+instruction generations may occasionally add a short extra phrase near the end of the clip. This behavior is also reproducible in the upstream Python path with the same reference/text/seed, so it is treated as a current v4 model/runtime limitation rather than a GGUF-only issue. No-reference and instruction-only paths are usually cleaner; for reference-conditioned use, try a different seed, instruction, or explicit duration_sec if the tail matters.

Quick Start

No-reference v4 speech:

audiocpp_cli --task tts --family irodori_tts \
  --model models/Irodori-TTS-v4-Small-GGUF/irodori-tts-v4-small-q8_0.gguf \
  --backend cuda --language ja \
  --text "今日は短い確認です。やさしく、聞き取りやすい声でお願いします。" \
  --request-option no_ref=true \
  --out out.wav

v4 voice cloning:

audiocpp_cli --task clon --family irodori_tts \
  --model models/Irodori-TTS-v4-Small-GGUF/irodori-tts-v4-small-q8_0.gguf \
  --backend cuda --language ja \
  --text "どうしてもっと早く教えてくれなかったの？私、ずっと待ってたのに。" \
  --voice-ref models/Irodori-TTS-v4-Small/samples/clone_ref1.wav \
  --out out.wav

v4 voice design:

audiocpp_cli --task vdes --family irodori_tts \
  --model models/Irodori-TTS-v4-Small-GGUF/irodori-tts-v4-small-q8_0.gguf \
  --backend cuda --language ja \
  --text "本日はお越しいただき、誠にありがとうございます。" \
  --request-option instruction="落ち着いた大人の男性。深く響く声で丁寧に話している。" \
  --request-option no_ref=true \
  --request-option guidance_scale=3 \
  --out out.wav

v3 voice cloning:

audiocpp_cli --task clon --family irodori_tts \
  --model models/Irodori-TTS-500M-v3-GGUF/irodori-tts-500m-v3-q8_0.gguf \
  --backend cuda --language ja \
  --text "同じ声で短く話します。" \
  --voice-ref models/Irodori-TTS-500M-v3/samples/clone_ref1.wav \
  --request-option no_ref=false \
  --out out.wav

Request Options

v4 uses the normalized schema-v1 option names directly. New requests should use these names:

Option	Values	Default	Meaning
language	ja	ja	Text language code; Irodori-TTS is Japanese-only.
instruction	text	empty	Voice-design instruction; only useful on caption-conditioned checkpoints. Legacy caption is accepted as an alias.
no_ref	bool	true unless a reference is provided	Use no-reference generation. Set false with --voice-ref for reference conditioning.
num_inference_steps	integer	40	RF diffusion steps.
duration_sec	seconds	unset	Explicit output duration; omitted uses predicted duration.
duration_scale	float	1.0	Multiplier for predicted duration.
min_duration_sec	seconds	0.5	Minimum generated duration.
max_duration_sec	seconds	30	Maximum generated duration.
text_chunk_mode	japanese, endline	endline	Long-form chunking mode.
text_chunk_size	integer	model text window	Maximum characters per outer chunk.
text_guidance_scale	float	3.0	Text CFG strength.
speaker_guidance_scale	float	5.0	Speaker CFG strength.
caption_guidance_scale	float	3.0	Caption CFG strength.
guidance_scale	float	unset	Override all CFG strengths when set.
guidance_mode	independent, joint, alternating	independent	CFG combination mode.
guidance_min_t	float	0.5	Minimum diffusion timestep where guidance is active.
guidance_max_t	float	1.0	Maximum diffusion timestep where guidance is active.
seed	integer	random	Generation seed.
trim_tail	bool	true	Trim trailing silence-like samples.

Session Options

Option	Values	Default	Meaning
irodori_tts.weight_type	native, f32, f16, bf16, q8_0	native	Model weight storage type.
irodori_tts.codec_weight_type	native, f32, f16, q8_0	native	DACVAE codec weight storage type.
irodori_tts.mem_saver	bool	true	Release staged runtime graphs after request phases.
irodori_tts.reference_cache_slots	integer	1	Prepared reference-speaker cache slots; use 0 to disable reuse.
irodori_tts.condition_graph_arena_mb	MiB	256	Condition encoder graph arena size.
irodori_tts.rf_graph_arena_mb	MiB	768	RF sampler graph arena size.
irodori_tts.codec_graph_arena_mb	MiB	512	DACVAE codec graph arena size.
irodori_tts.condition_weight_context_mb	MiB	32	Condition encoder weight metadata context size.
irodori_tts.rf_weight_context_mb	MiB	32	RF sampler weight metadata context size.
irodori_tts.codec_weight_context_mb	MiB	32	DACVAE codec weight metadata context size.

Compatibility

The runtime accepts the old option names below for existing local scripts and older standalone GGUF packages. Prefer the v1 names for new requests.

Legacy option	v1 option
caption	instruction
duration_seconds	duration_sec
min_seconds	min_duration_sec
max_seconds	max_duration_sec
mem_saver	irodori_tts.mem_saver
reference_cache_slots	irodori_tts.reference_cache_slots