cstr/pocket-tts-GGUF

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, 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 (pocket-tts-english-{f16,q8_0,q4_k}.gguf — the ones without novc in the name) embed the Mimi VAE encoder and speaker projection needed to condition on a reference; the 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

Two families: voice-cloning (default, embeds the Mimi encoder) and novc (decoder only, smaller — use when you only need the built-in default voice).

File	Voice clone	Quant	Size	Notes
pocket-tts-english-f16.gguf	✅	F16	219 MB	Reference quality, cloning
pocket-tts-english-q8_0.gguf	✅	Q8_0	124 MB	Near-F16, cloning
pocket-tts-english-q4_k.gguf	✅	Q4_K	73 MB	Smallest with cloning
pocket-tts-english-novc-f16.gguf	—	F16	200 MB	Decoder only
pocket-tts-english-novc-q8_0.gguf	—	Q8_0	110 MB	Decoder only
pocket-tts-english-novc-q4_k.gguf	—	Q4_K	62 MB	Decoder only, smallest

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 (pulls the voice-cloning F16):

./build/bin/crispasr -m pocket-tts --auto-download \
    --tts "The quick brown fox jumps over the lazy dog." \
    --tts-output fox.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.

Provenance and EU AI Act Art. 53 note

• Upstream model: kyutai/pocket-tts — published by kyutai.
• Upstream licence: cc-by-4.0. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF/GGML). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository. No training-content summary was found on the upstream model card at the time of writing; that documentation gap is upstream's and is not filled here.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.