Pocket TTS — GGUF (ggml-quantised)

GGUF / ggml conversion of kyutai/pocket-tts for use with CrispStrobe/CrispASR.

Pocket TTS is a lightweight (~100M param) continuous-latent autoregressive TTS model from Kyutai, based on the CALM paper (arXiv:2509.06926). Unlike codebook-based TTS models, Pocket TTS emits continuous float vectors — no discrete tokens, no softmax sampling:

• FlowLM backbone — causal transformer (1024D, 16 heads, 6 layers; 24 layers for the French preview, RoPE, GELU) operating at 12.5 Hz
• Consistency head — SimpleMLPAdaLN (512D, 6 ResBlocks) with timestep embedding → one-step LSD decode → 32-dim continuous latent vectors
• Mimi VAE decoder — SEANet upsample convolutions + 2-layer transformer → 24 kHz PCM
• Mimi VAE encoder (voice-cloning builds only) — SEANet downsample + 2-layer transformer + speaker projection → reference conditioning
• Text tokenizer — SentencePiece BPE (4000 vocab, embedded in GGUF)

Single GGUF file — no separate codec companion needed (Mimi weights and the tokenizer are embedded).

Released under CC-BY-4.0 license.

Voice cloning

Pocket TTS is zero-shot: it clones the timbre of a short reference clip. The voice-cloning builds (the files without novc in the name) embed the Mimi VAE encoder and speaker projection needed to condition on a reference; the English-only novc builds omit that encoder and are ~20 MB smaller.

# clone the timbre of ref.wav (any sample rate; mono is used)
./build/bin/crispasr --backend pocket-tts -m pocket-tts-english-f16.gguf \
    --voice ref.wav \
    --tts "The quick brown fox jumps over the lazy dog." \
    --tts-output fox.wav --seed 42

Pocket TTS produces near-silence without voice conditioning, so if you omit --voice a built-in default reference is used automatically. Use a clean, single-speaker reference of a few seconds for best results.

Files

The multilingual Q8_0/F16 files are full voice-cloning models. German, Spanish, Italian, and Portuguese are Kyutai's distilled 6-layer releases; French is the larger, undistilled 24-layer preview. The decoder-only novc alternatives are currently English-only.

F16 equivalents of every multilingual Q8_0 file are also available (about 219 MB for each 6-layer model and 673 MB for French).

Quick start

# 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-cli

# 2. Download a model (voice-cloning F16 shown)
huggingface-cli download cstr/pocket-tts-GGUF pocket-tts-english-f16.gguf --local-dir .

# 3. Synthesize
./build/bin/crispasr --backend pocket-tts -m pocket-tts-english-f16.gguf \
    --tts "Hello, how are you today?" \
    --tts-output hello.wav --seed 42

Or with auto-download. -l de, es, it, pt, or fr selects and caches the matching Q8_0 checkpoint; omit -l for English:

./build/bin/crispasr --backend pocket-tts -m auto --auto-download -l es \
    --accept-license pocket-tts-terms \
    --voice ref.wav --i-have-rights \
    --tts "Hola, este modelo ya habla español." \
    --tts-output hola.wav

Python binding

from crispasr import Session

sess = Session("pocket-tts-english-f16.gguf")
sess.set_tts_seed(42)
pcm = sess.synthesize("Hello world.")
sess.write_wav("hello.wav", pcm)

Conversion

Converted with models/convert-pocket-tts-to-gguf.py from the CrispASR repo (--voice-cloning bakes in the Mimi encoder + speaker projection). The Mimi codec and SentencePiece tokenizer are embedded in the single GGUF.