cielvox2-tts-1.7b-voicedesign-GGUF

Xenna/cielvox2-tts-1.7b-voicedesign-GGUF

cielvox2-tts-1.7b-voicedesign-GGUF is a text-to-speech model from Xenna. Use it when you need text read aloud. It is set up for ggml. The card lists the license as apache-2.0.

Local-first fixed-speaker TTS model for the cielvox2-tts-1.7b-customvoice backend in stelnetxcis-create/cielvox2.

Downloads · 30 days

15% of all-time downloads

All-time downloads

181

Public

Repo size

2 GB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.gguf2 GB · 100%

At a glance

Task: Text-to-Speech
Library: ggml
License: apache-2.0
Access: Public
Created: Aug 21, 2026
Updated: Aug 22, 2026
SHA: 52dc55c3

Model card

From the Hugging Face model README

CielVox 1.7B CustomVoice — GGUF

Local-first fixed-speaker TTS model for the cielvox2-tts-1.7b-customvoice backend in stelnetxcis-create/cielvox2.

9 baked speaker tokens picked via --voice <name>
No ECAPA forward, no codec encoder, no reference audio required
Two speakers (dylan, eric) carry Chinese-dialect overrides (Beijing / Sichuan)
Apache-2.0 licence

Pair this with the codec at Xenna/cielvox2-tokenizer-12hz — the talker emits 16-codebook RVQ codes that the codec decoder renders to 24 kHz PCM.

Speaker	Language / dialect
`aiden` (default)	English (M)
`dylan`	Beijing dialect (M, dialect_token=2074)
`eric`	Sichuan dialect (M, dialect_token=2062)
`ono_anna`	English (F)
`ryan`	English (M)
`serena`	English (F)
`sohee`	English (F)
`uncle_fu`	English (M, older)
`vivian`	English (F)

Files

File	Quant	Size	Notes
`cielvox2-tts-12hz-1.7b-customvoice-q8_0.gguf`	Q8_0	2.04 GB	Recommended

Quick start

# 1. Build
git clone https://github.com/stelnetxcis-create/cielvox2
cd cielvox2
cmake -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build -j --target stelnettts-lib

# 2. Pull the talker + codec
huggingface-cli download Xenna/cielvox2-tts-1.7b-customvoice-GGUF \
    cielvox2-tts-12hz-1.7b-customvoice-q8_0.gguf --local-dir .
huggingface-cli download Xenna/cielvox2-tokenizer-12hz \
    cielvox-tokenizer-12hz.gguf --local-dir .

# 3. Synthesise — pick a speaker by name
./build/bin/stelnettts --backend cielvox2-tts-1.7b-customvoice \
    -m cielvox2-tts-12hz-1.7b-customvoice-q8_0.gguf \
    --codec-model cielvox-tokenizer-12hz.gguf \
    --voice ryan \
    --tts "Hello, this is the Ryan speaker on the 1.7B model." \
    --tts-output ryan.wav

For auto-download simply pass -m auto:

./build/bin/stelnettts --backend cielvox2-tts-1.7b-customvoice -m auto \
    --voice vivian \
    --tts "Auto-download fetches both files." \
    --tts-output out.wav

Quality verification

Speaker	Output (verbatim)
`ryan`	`"Hello, this is the Ryan speaker."`
`vivian`	`"This is a longer prompt to verify the 1.7B CustomVoice path through the talker."`

Both roundtrips are exact-match on a long prompt.

Architecture

Component	Details
Talker LM	28 layers, 2048 hidden, 16 heads, 8 KV heads, head_dim=128
Output head	16 codebooks × 2048 (RVQ)
`small_to_mtp_projection`	2048-d → 1024-d MLP (input bridge to code predictor) — applied per-step at decode
Code predictor	5L + 15 separate codec_embedding/lm_head pairs (top-k=50, temp=0.9)
Codec	CielVox-Tokenizer-12Hz (separate GGUF, 12.5 fps RVQ)
Audio	24 kHz mono float32 PCM

Implementation

The C++ runtime is implemented in src/cielvox2_tts.cpp using ggml graphs for the talker / code-predictor path and the companion tokenizer GGUF for codec encode/decode.

Attribution

GGUF conversion + ggml runtime: stelnetxcis-create/cielvox2 — see src/cielvox2_tts.cpp, models/convert-cielvox2-tts-to-gguf.py.

License

Apache 2.0.

Task: Text-to-Speech
Library: ggml
License: apache-2.0
Languages: en, zh, de, fr, it, es
Created: Aug 21, 2026
Updated: Aug 22, 2026