Re-export qwen3-tts-12hz-0.6b (loom-exporter 1891005)

README.md

@@ -50,24 +50,19 @@ pip install -U "loom-py-rt[hub]"
 
 ```python
 import loom
-import numpy as np
+import librosa
 
 model = loom.Model.from_pretrained("loom-ai-org/qwen3-tts-12hz-0.6b-loom")
 
-# The voice is an X-VECTOR -- 1024 floats -- not a speaker id: this checkpoint carries no speaker
-# table. Making one from a reference clip inside loom is temporarily withdrawn (see "Known
-# limitations", which also gives the upstream recipe that wrote this file).
-x_vector = np.loadtxt("x_vector.txt", dtype=np.float32)
-
-# The text goes in the prompt format this model was trained on, spelled out: the file declares no
-# chat template, and the bare sentence makes it stop early or ramble (see "Known limitations").
-role = "<|im_start|>assistant\n"
+# The voice is a REFERENCE CLIP, not a speaker id -- this checkpoint carries no speaker table. A few
+# clear seconds is enough. 24 kHz is what the speaker encoder expects, so resample on the way in.
+reference, _ = librosa.load("reference.wav", sr=24000)
 
 # What comes back is codec TOKENS, not audio -- frame-major, one row per frame, 16 codebooks wide.
 # `max_new_tokens` counts AUDIO FRAMES, at 12.5 per second.
 codes = model.text2codes.infer(
-    f"{role}The quick brown fox jumps over the lazy dog.<|im_end|>\n{role}",
-    x_vector=x_vector.tolist(),
+    "The quick brown fox jumps over the lazy dog.",
+    waveform=reference.tolist(),
     language_id=2050,          # English; see "Known limitations" for the rest
     max_new_tokens=200, seed=1234,
 )
@@ -105,11 +100,11 @@ accepts for this model, and is the authority on it. See [loom-py](https://github
 
 **This model does not produce audio.** It emits sixteen streams of Qwen3-TTS codec tokens, and the codec turns those into a waveform -- [`qwen3-tts-tokenizer-12hz-loom`](https://huggingface.co/loom-ai-org/qwen3-tts-tokenizer-12hz-loom), which ships inside this same checkpoint upstream. They stay separate because one codec serves every size and variant of this talker, and because the codes are the useful intermediate.
 
-**The voice is an x-vector, and there is no speaker table.** This checkpoint's `spk_id` is empty, so cloning is the only mode: pass `x_vector=`, 1024 floats that are the whole of what the voice contributes. **Making one from a clip inside loom is temporarily withdrawn:** this file's speaker encoder (`waveform=`) currently aborts the process -- a native assertion, not a Python exception -- so do not pass `waveform=` until a release notes the fix. Until then make the x-vector once with the upstream `qwen-tts` package and save it: `spk = Qwen3TTSForConditionalGeneration.from_pretrained("Qwen/Qwen3-TTS-12Hz-0.6B-Base").extract_speaker_embedding(audio=wav, sr=24000)` on a few clear seconds of the voice resampled to 24 kHz, then `np.savetxt("x_vector.txt", spk.float().numpy().reshape(-1))`. The same file serves every later call.
+**The voice comes from a reference clip, and there is no speaker table.** This checkpoint's `spk_id` is empty, so cloning is the only mode: pass `waveform=` and the speaker encoder extracts an x-vector from it. Pass `x_vector=` instead to reuse one you already have -- it is 1024 floats and it is the whole of what the voice contributes. Copies of this file downloaded before 2026-09-29 abort the process on `waveform=`; download it again.
 
-**Pass the text in the model's prompt format, as the example does.** The talker was trained on `<|im_start|>assistant\n{text}<|im_end|>\n<|im_start|>assistant\n`, and this file does not declare that template, so `text2codes` sends exactly what you give it. Spelled out, the prompt tokenizes to upstream's ids and a greedy run (`temperature=0, subtalker_temperature=0`) matches `transformers` on every code; the bare sentence makes the model stop after a word or two, or run to `max_new_tokens`.
+**Pass the sentence as it is.** The talker was trained on `<|im_start|>assistant\n{text}<|im_end|>\n<|im_start|>assistant\n`, and this file wraps whatever it is given in that template. Ids you have already wrapped yourself -- `tokens=` that open with `<|im_start|>assistant\n` -- are passed through as they are, and so is a reference transcript in ICL mode below.
 
-**The reference TEXT is used if you pass it -- that is ICL mode.** Upstream's higher-fidelity clone mode conditions on a transcript of the reference clip plus the clip's own codec tokens, replayed as if the model had just spoken it. Pass `ref_audio=` (the same clip, 24 kHz) and `ref_tokens=` (its transcript, tokenized with this model's own tokenizer and template) alongside `x_vector=`, and this file draws those codes itself: the codec's ENCODE half rides inside this GGUF, since the prompt is its only caller, and it is unaffected by the speaker encoder's abort. Pass `ref_code=` instead if you already hold them, frame-major and sixteen wide. Omit both and you get the x-vector mode above, which is what the rest of this card describes.
+**The reference TEXT is used if you pass it -- that is ICL mode.** Upstream's higher-fidelity clone mode conditions on a transcript of the reference clip plus the clip's own codec tokens, replayed as if the model had just spoken it. Pass `ref_audio=` (the same clip, 24 kHz) and `ref_tokens=` (its transcript, tokenized with this model's own tokenizer) and this file draws those codes itself: the codec's ENCODE half rides inside this GGUF, since the prompt is its only caller. Pass `ref_code=` instead if you already hold them, frame-major and sixteen wide. Omit both and you get the x-vector mode above, which is what the rest of this card describes.
 
 **A reference clip is used in whole codec frames.** 1920 samples at 24 kHz, 80 ms: the driver trims to a multiple of that before encoding, so up to 79 ms of the clip's tail is not heard. Every convolution in the encode stack pads by a length-derived amount that would not otherwise trace, and on a frame boundary that amount is provably zero.
 

qwen3-tts-12hz-0.6b.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5871b9bc7803465b0a7e90ce8a1bbee2887214fd9ab3474ef75808037276b0ca
-size 4132302976
+oid sha256:ceb1bb8e543d2eacc82b9892fef2ab4746891a1e8d0c901c0ab8024b82c75540
+size 4132304416