borisbat/dasllama-tts

dasLLAMA text-to-speech models

The speech models dasLLAMA serves, converted to GGUF, and the two front-end packs the phoneme families load. dasLLAMA is the daslang inference engine (https://github.com/GaijinEntertainment/daScript, modules/dasLLAMA); the models are the StyleTTS2 lineage - KittenTTS nano and mini, Kokoro-82M - behind one facade and one das-native text front end (a normalizer, a part-of-speech tagger, a grapheme-to-phoneme pass), and Kyutai's Pocket TTS, a continuous-audio language model that reads text and clones a voice from a clip. No espeak-ng, no phonemizer: the front end is data, and the data is in the two packs.

Files

file	what	bytes	sha256
kitten-nano.gguf	KittenTTS nano 0.8, f32	59331456	4556948c36a29e4be5ad521e597a20ae817059c404fea3c5d935afa73506d9da
kitten-mini.gguf	KittenTTS mini 0.8, f32	295975008	e127a95c6ffa390eccc5956e7156ff055e9f30078cf16361cbcd0eea6c21fdb5
kokoro-82m.gguf	Kokoro-82M v1.0 with its 54 voice packs, f32	352965024	e0d6584b5b650730bd62e7b105a2e9502384bbe3fbebf3030113560355e6232f
tts_g2p.bin	the grapheme-to-phoneme pack: misaki's gold and silver lexicons in both English dialects (one merged table per tier), CMUdict 0.7a rendered into the American inventory, the g2p_en GRU spelling model	14011554	d7b6afea7a0901a877d10531054d3f967845c58f415f0ea937decd8a3081ccd5
tts_g2p_en_us.bin	the same pack with the American tier alone - no British values, CMUdict pruned of every word the American lexicon carries; the web serving set's pack, British voices are not offered on it	10257455	6f69d2e74565bd7d876b8d1f4042bf8c1c5b615387fa26ff45215cf447932154
tts_postag.bin	the tokenizer exception table and the averaged-perceptron PTB tagger	12566510	38c2e85f7fef3e57d561d2aa0af25fccda4276376ba1993c3dbc2ae0ebfa57b4
pocket-tts-en-q8.gguf	Kyutai Pocket TTS, English (english_2026-04): the served GEMM weights as Q8_0, the rest f16, its unigram tokenizer and 19 voice clips inside; reads text, needs no pack	152613664	9fca82dbe1a550a0bce958d3dfcf51f3c793c89db6aee850823724826b56589c
pocket-tts-de-q8.gguf	Pocket TTS German (6 layers), the same form, one voice (juergen)	134667200	a7f69bff844d796a164a62766071457f56c5bd8d854a0a743f88c2d8c4790d9e
pocket-tts-es-q8.gguf	Pocket TTS Spanish (6 layers), one voice (lola)	134624480	40b36e28cbc1d6d01ef660751b63b37b44f25b6887a93102bc29bfb516e93ade
pocket-tts-it-q8.gguf	Pocket TTS Italian (6 layers), one voice (giovanni)	134415072	3c5739d544b1b7c8284fd3df9d7122557cf700c91895d3dc45b3fdc5ef6e2670
pocket-tts-pt-q8.gguf	Pocket TTS Portuguese (6 layers), one voice (rafael)	134667488	3375c31e742c8783c6dddbbd3bd152e8dff9d188fdaceb1cbb4d187514291c57
pocket-tts-fr-q8.gguf	Pocket TTS French (24 layers, the only French model Kyutai ships), one voice (estelle)	375793696	f06ffac80b96a34d2e51ca40c41111469d8b44e0269b27a64e707a7a9be1ec20
pocket-tts-en-kq.gguf	Pocket TTS English in the small form: the backbone and the codec transformers as Q4_K, the flow head and the codec convolutions as Q8_0, the embedding table Q4_K; its tokenizer, the codec encoder (so it clones) and the 19 voices as latent frames	74970016	2475a1ed8d49eb72c9d9b8c38f10f91ef5b03c7cd6e9fe43fdf7ab00ae1a0a25
pocket-tts-en-stuart-kq.gguf	the same small form with one voice (stuart_bell) as latent frames and no codec encoder: reads text in that voice, cannot clone	65107520	bc9604b527066134354dc480e20c960f63f5c3538c1dd757ba409bd782cddac9

The packs sit beside whichever GGUF you load; the loader reads them from the model's directory - tts_g2p.bin when it is there, else tts_g2p_en_us.bin. The GGUFs carry f32 weights: dasLLAMA quantizes the served layouts to Q8_0 at first load and keeps the result beside the file as a prepared image, so the f32 file is also the reference lane.

Use

dasllama-server --tts kitten-nano.gguf          # POST /v1/audio/speech, the OpenAI shape
daslang utils/dasllama-server/txt2wav.das -- --tts kitten-nano.gguf --voice expr-voice-2-f --text "Hello." --out hello.wav

Pocket TTS English is the cloning model: 152 MB, 19 voices (alba the default, bill_boerst, caro_davy, anna, george, ...), and any voice from a few seconds of 24 kHz audio through tts_register_voice. It reads text, so it needs neither pack; the English normalizer runs in front of it. On the 200-sentence rig at alba this file reads WER 3.91 / UTMOS 4.328 at a real-time factor of 0.051 on an Apple M1 Max, against the reference package's 5.00 / 4.393 / 0.210 (measured 2026-09-09 with the module's harness/tts_rig.py, the engine under the JIT tier with the box's tune profile, the reference package under torch on one thread). The five other languages are the same form, one file each with Kyutai's default clip for that language as its only voice (German juergen, Spanish lola, Italian giovanni, Portuguese rafael, French estelle); the German, Spanish, Italian and Portuguese files are the six-layer models, French exists only as the 24-layer one. A voice cloned from any clip speaks the file's language with the clip's accent. Text in those languages is read as it is, since the normalizer is English. pocket-tts-en-kq.gguf is the English model in the small form, 75 MB: the backbone and the codec transformers as Q4_K, the flow head and the codec convolutions as Q8_0, the embedding table Q4_K, the encoder and the 19 voices inside (on the rig at alba: WER 3.86 / UTMOS 4.295 at a real-time factor of 0.049 on the same box); it is the file the browser examples on dasllama.io fetch. pocket-tts-en-stuart-kq.gguf is that form with one voice, stuart_bell, and no codec encoder, 65 MB: it reads text in that voice and cannot clone.

Kitten nano is the phoneme families' served default: 59 MB, eight voices, a real-time factor of 0.03 on an Apple M1 Max (measured 2026-09-02 with the same rig). Its voices are expr-voice-2-m through expr-voice-5-f, and the upstream's alias names (Bella, Jasper, ...) are accepted for them. Kokoro ships 54 packs, of which the front end drives the 28 English ones: the 20 American (af_*, am_* - af_heart, am_adam, ...) and the 8 British (bf_*, bm_* - bf_emma, bm_george, ...), each phonemized in its own dialect. The other 26 packs speak languages the front end does not phonemize yet, so they are not listed and are refused. Some Kokoro packs, the British ones most audibly, render a comma or a full stop as a breath; the reference pipeline does the same, so it is the pack, not the engine. Every model speaks at 24 kHz. GET /v1/stats on the server lists the served model's voices.

Provenance

• kitten-nano.gguf / kitten-mini.gguf: converted by modules/dasLLAMA/harness/convert_kitten.py from KittenML/kitten-tts-nano-0.8 at 7a1db645b1f3ab9420761d87428e042b9cec3f26 and KittenML/kitten-tts-mini-0.8 at c02725660cea441db4c383af69f1f26f5cd00947 (the ONNX weights, the voice table, the config).
• kokoro-82m.gguf: converted by modules/dasLLAMA/harness/convert_kokoro.py from hexgrad/Kokoro-82M at f3ff3571791e39611d31c381e3a41a3af07b4987 (kokoro-v1_0.pth and the voice packs).
• tts_g2p.bin: modules/dasLLAMA/harness/build_g2p_data.py over misaki 0.9.4 (us_gold.json, us_silver.json, gb_gold.json, gb_silver.json), CMUdict 0.7a (the NLTK distribution) and g2p_en 2.1.0 (checkpoint20.npz), plus harness/g2p_local_additions.json.
• tts_postag.bin: modules/dasLLAMA/harness/train_postag.py over UD English-EWT and Project Gutenberg prose tagged by spaCy's en_core_web_sm; the tokenizer exception table is spaCy's.
• pocket-tts-en-q8.gguf: converted by modules/dasLLAMA/harness/convert_pocket.py --q8 from kyutai/pocket-tts languages/english_2026-04/model.safetensors at 19f95fe2df36e79fbd9f10008595cc4c977a0fcc and the tokenizer of kyutai/pocket-tts-without-voice-cloning at d29db7978e464fb90cb3359ee0c69a273b9142cc; the voice clips from kyutai/tts-voices at 323332d33f997de8394f24a193e1a76df720e01a (voice-zero/, voice-donations/, vctk/, alba-mackenna/casual.wav).
• pocket-tts-en-kq.gguf / pocket-tts-en-stuart-kq.gguf: the same sources through modules/dasLLAMA/harness/convert_pocket.py --kq (the second with --voices stuart_bell --no-cloning); the K-quant blocks are ggml's own quantizer, and each voice is stored as the latent frames of its clip through the model's codec encoder.

The whole set is rebuilt by modules/dasLLAMA/performance/build_tts_data.das. Parity against the reference implementations (block by block, and the front end sentence by sentence on a 200-sentence corpus) is the test suite under modules/dasLLAMA/tests/test_tts_*.das.

Licences, per file

file	licence	terms
kitten-nano.gguf, kitten-mini.gguf	Apache-2.0	KittenML's weights, converted; LICENSE.APACHE-2.0; the architecture is StyleTTS2 (MIT, LICENSE.STYLETTS2)
kokoro-82m.gguf	Apache-2.0	hexgrad's weights and voices, converted; the architecture is StyleTTS2 (MIT, LICENSE.STYLETTS2)
tts_g2p.bin	Apache-2.0 and BSD-2-Clause	misaki and g2p_en (Apache-2.0), CMUdict (LICENSE.CMUDICT, Carnegie Mellon University)
tts_postag.bin	CC BY-SA 4.0	the tagger weights are trained on UD English-EWT (LICENSE.UD_EWT); the exception table and the silver tags come from spaCy (MIT, LICENSE.SPACY); Gutenberg prose is public domain
pocket-tts-*.gguf (every Pocket file, the two -kq ones included)	CC BY 4.0	Kyutai's weights and tokenizer, converted (LICENSE.CC-BY-4.0); the reference implementation is MIT (LICENSE.POCKET_TTS) and not included; the voice clips: voice-zero and voice-donations CC0, VCTK (CSTR, University of Edinburgh) and Alba Mackenna CC BY 4.0 - the sidecar lists each

Each .LICENSE sidecar beside a file names its sources; the full texts are in this repository. The engine that reads these files is under the daslang licence in its own repository.