Downloads · 30 days
296
24% of all-time downloads
JJarvinen/chatterbox-finnish-nano-GGUF
chatterbox-finnish-nano-GGUF is a text-to-speech model from JJarvinen. Use it when you need text read aloud. It is set up for ggml. The card lists the license as mit.
GGUF conversion of JJarvinen/chatterbox-finnish-nano v0.1.3 for CrispASR. This is the Finnish-only Nano T3 release. The built-in no-reference Nano voice conditioning is embedded in the T3 GGUF.
Downloads · 30 days
296
24% of all-time downloads
All-time downloads
1.2K
Public
Repo size
1.1 GB
Likes
0
Public
Click a slice to open those files.
.gguf1.1 GB · 100%
From the Hugging Face model README
GGUF conversion of JJarvinen/chatterbox-finnish-nano v0.1.3 for CrispASR.
This is the Finnish-only Nano T3 release. The built-in no-reference Nano voice
conditioning is embedded in the T3 GGUF.
| File | Size | SHA256 | Notes |
|---|---|---|---|
chatterbox-finnish-nano-v0.1.3-t3-f16.gguf | 453.7 MiB | 24b99c1ec01339c34a4334aa06f70ab97bbbb455421ee71389bafbbf112b4d1e | Reference conversion |
chatterbox-finnish-nano-v0.1.3-t3-q8_0.gguf | 329.2 MiB | f804e82ada51c848e68b3f98973a7e208cf7e06432590fc14b6fb7777826f6a2 | Recommended quality/size default |
chatterbox-finnish-nano-v0.1.3-t3-q4_k.gguf | 270.2 MiB | 1683dc906b291866a2f12368822e2f140ed3b39e58231c8d31ee25f526fe5ae8 | Smaller deployment build |
Chatterbox Nano uses the same MeanFlow S3Gen, VE and conditioning assets as Chatterbox Turbo. The custom Finnish weights are in T3, so this repository does not duplicate the unchanged S3Gen GGUF.
Download a companion from cstr/chatterbox-turbo-GGUF, for example Q8_0:
hf download cstr/chatterbox-turbo-GGUF \
chatterbox-turbo-s3gen-q8_0.gguf \
--local-dir .
Available companion files there include:
chatterbox-turbo-s3gen-f16.ggufchatterbox-turbo-s3gen-q8_0.ggufchatterbox-turbo-s3gen-q4_k.gguf./build/bin/crispasr \
--backend chatterbox-nano \
--no-gpu \
-m chatterbox-finnish-nano-v0.1.3-t3-q4_k.gguf \
--codec-model chatterbox-turbo-s3gen-q8_0.gguf \
--tts "Hyvää huomenta. Tämä on Chatterbox Finnish Nano." \
--tts-output out.wav
--no-gpu is used above because it is also the release smoke-test path and
separates model/GGUF validation from GPU-backend-specific behavior. Consult the
current CrispASR documentation for the latest GPU support status.
Source model revision:
JJarvinen/[email protected]
Canonical source FP16 t3_nano_v1.safetensors SHA256:
d25e3ac95eefefb58c40ad4d5a47b4fa621bc1bd92e91f2d904f622d5458ad26
Converter:
CrispStrobe/CrispASR
commit 3b4ed4fb83767fa098336f673804db7294f5da5c
models/convert-chatterbox-to-gguf.py --variant nano
F16 was converted first. Q8_0 and Q4_K were then produced with CrispASR's
model-agnostic crispasr-quantize, which preserves non-quantizable and
architecture-guarded tensors at their required precision.
The release helper performs CPU synthesis smoke tests (--no-gpu) for all
three T3 variants before publishing. The test uses the unchanged Turbo Q8_0
S3Gen companion. Listen to the generated WAVs before running the publisher.
This GGUF conversion does not change the v0.1.3 model architecture, vocabulary, training data or intended capabilities; it changes the serialized numeric representation for the CrispASR/ggml runtime.