Kartoffel Orpheus 3b German Natural GGUF
GGUF / ggml conversion of SebastianBodza/KartoffelOrpheus-3Bgermannatural-v0.1 for use with CrispStrobe/CrispASR.
Fuente del modelo
- Tipo
- Other
- Tamaño del archivo
- 1,74–6,16 GB
- Modelo base
- SebastianBodza/Kartoffel_Orpheus-3B_german_natural-v0.1
- Versión
- vRevision 36eca4d
- Actualizado
- 26 sept 2026
- Descargas
- 582
- Fuente
- Hugging Face
Extracto de la fuente
GGUF / ggml conversion of SebastianBodza/KartoffelOrpheus-3Bgermannatural-v0.1 for use with CrispStrobe/CrispASR.
Fuentes
1 fuente
Fuentes
1 fuenteVerificado 6 sept
Artefactos del modelo
3 artefactos
Artefactos del modelo
3 artefactoskartoffel-orpheus-de-natural-f16.gguf
gguf · 6,16 GB · SHA-256 d2c091285db4…30de · Hugging Face
Descargarkartoffel-orpheus-de-natural-q4_k.gguf
gguf · 1,74 GB · SHA-256 cac11bae7ce5…b1a3 · Hugging Face
Descargarkartoffel-orpheus-de-natural-q8_0.gguf
gguf · 3,27 GB · SHA-256 f8f5061307c3…fc7d · Hugging Face
DescargarExtractos de fuentes
2 extractos
Extractos de fuentes
2 extractos--- license: llama3.2 language: - de base_model: - SebastianBodza/Kartoffel_Orpheus-3B_german_natural-v0.1 - canopylabs/orpheus-3b-0.1-ft - meta-llama/Llama-3.2-3B-Instruct pipeline_tag: text-to-speech tags: - tts - text-to-speech - orpheus - kartoffel - llama - snac - german - deutsch - gguf - crispasr library_name: ggml --- # Kartoffel-Orpheus 3B (German, natural) — GGUF (ggml-quantised) GGUF / ggml conversion of [`SebastianBodza/Kartoffel_Orpheus-3B_german_natural-v0.1`](https://huggingface.co/SebastianBodza/Kartoffel_Orpheus-3B_german_natural-v0.1) for use with **[CrispStrobe/CrispASR](https://github.com/CrispStrobe/CrispASR)**. A German fine-tune of [`canopylabs/orpheus-3b-0.1-ft`](https://huggingface.co/canopylabs/orpheus-3b-0.1-ft), trained primarily on **natural human speech recordings** from German speakers. Drop-in checkpoint swap on the same Orpheus runtime — same Llama-3.2-3B-Instruct talker arch, same SNAC 24 kHz codec, same `<custom_token_N>` super-frame protocol, just different LM weights and a German speaker roster. The synthetic-data sibling lives at [`cstr/kartoffel-orpheus-3b-german-synthetic-GGUF`](https://huggingface.co/cstr/kartoffel-orpheus-3b-german-synthetic-GGUF). | Speakers | Names | |---|---| | Male | `Jakob`, `Anton`, `Julian`, `Jan`, `Alexander`, `Emil`, `Ben`, `Elias`, `Felix`, `Jonas`, `Noah`, `Maximilian` | | Female | `Sophie`, `Marie`, `Mia`, `Maria`, `Sophia`, `Lina`, `Lea` | Pair this with the SNAC codec at [`cstr/snac-24khz-GGUF`](https://huggingface.co/cstr/snac-24khz-GGUF) — the talker outputs codec tokens but doesn't render audio without it. ## Files | File | Quant | Size | Notes | |---|---|---:|---| | `kartoffel-orpheus-de-natural-f16.gguf` | F16 | ~6.2 GB | Reference quality | | `kartoffel-orpheus-de-natural-q8_0.gguf` | Q8_0 | ~3.4 GB | **Recommended** | | `kartoffel-orpheus-de-natural-q4_k.gguf` | Q4_K | ~1.8 GB | Smallest; smoke-tested but Q8_0 preferred | Like the upstream Orpheus, sub-Q8 quants tend to break the SNAC super-frame slot pattern on rare prompts, so we ship Q4_K with a recommendation to use Q8_0 by default. ## Quick start ```bash # 1. Build CrispASR git clone https://github.com/CrispStrobe/CrispASR cd CrispASR cmake -B build -DCMAKE_BUILD_TYPE=Release cmake --build build -j --target crispasr-lib # 2. Pull the talker + the SNAC codec huggingface-cli download cstr/kartoffel-orpheus-3b-ger...
Source context: 582 downloads · 0 likes · Pipeline text-to-speech · Library ggml · Repo cstr/kartoffel-orpheus-3b-german-natural-GGUF
